You are on page 1of 22

8

8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


GENERATING ULTRAFILTERS IN A REASONABLE WAY
ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
Abstract. We continue investigations of reasonable ultralters on uncount-
able cardinals dened in Shelah [8]. We introduce a general scheme of generat-
ing a lter on from lters on smaller sets and we investigate the combinatorics
of objects obtained this way.
0. Introduction
Reasonable ultralters were introduced in Shelah [8] in order to suggest a line
of research that would in some sense repeat the beautiful theory created around
the notion of Ppoints on . The denition of reasonable ultralters involves two
conditions. The rst one, so called the weak reasonability of an ultralter, is a way
to guarantee that we are not entering the realm of large cardinals: the considered
ultralter is required to be very non-normal. Since this property will be also of some
interest in the present paper, let us recall the following denition and observation.
Denition 0.1 (Shelah [8, Def. 1.4]). (1) We say that a uniform ultralter D
on is weakly reasonable if for every non-decreasing unbounded function
f

there is a club C of such that
_
[, +f()) : C / D.
(2) Let D be an ultralter on , C be a club and let

: < ) be the
increasing enumeration of C 0. We dene
D/C =
_
A :
_
A
[

,
+1
) D
_
.
(It is an ultralter on .) D/C will be called the quotient of D by C.
Observation 0.2 (Shelah [8, Obs. 1.5]). Let D be a uniform ultralter on a regular
uncountable cardinal . Then the following conditions are equivalent:
(A) D is weakly reasonable,
(B) for every increasing continuous sequence

: < ) there is a club


C

of such that
_
_
[

,
+1
) : C

_
/ D,
Date: September 2007.
1991 Mathematics Subject Classication. Primary 03E05; Secondary: 03E20.
The rst author would like to thank the Hebrew University of Jerusalem and the Lady Davis
Fellowship Trust for awarding him with Schonbrunn Visiting Professorship under which this
research was carried out.
Both authors acknowledge support from the United States-Israel Binational Science Foundation
(Grant no. 2002323). This is publication 889 of the second author.
1
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


2 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
(C) for every club C of the quotient D/C does not extend the lter generated
by clubs of .
The second part of the denition of reasonable ultralters is directly related to
generalizing Ppoints to the context of weakly reasonable ultralters on an un-
countable cardinal . To carry out this process we have to be somewhat creative in
re-interpreting the property that any countable family of members of the ultralter
has a pseudo-intersection in the ultralter. An interesting way of doing this is to
impose some demands on how the ultralter on can be obtained from sequences
of objects on smaller cardinals (an approach motivated by Roslanowski and Shelah
[3, 5, 6].) For instance we may consider sequences r = (

, d

) : < ) such that

: < ) is an increasing continuous sequence of ordinals below and d

is an
ultralter on the interval [

,
+1
). For each such sequence r we look at the family
of subsets of which are eventually large in every interval [

,
+1
), that is we
consider the set l(r) =
_
A : ( < )( > )(A[

,
+1
) d

)
_
. (The set
l(r) is a lter on .) There is a natural quasi-order on sequences r as above: we say
that r

s if and only if l(r) l(s). Now the demand generalizing Ppointness


may be phrased for an ultralter D on as follows: there is a (<
+
)directed (with
respect to

) family H such that D =

l(r) : r H.
The ideas described above can be further generalized by allowing the use of
arbitrary lters or just lters with some special properties in place of ultralters
d

. This immediately leads to the general scheme of generating lters on presented


in our Denition 1.2.
This paper continues Shelah [8] and Roslanowski and Shelah [4], but it is essen-
tially self contained.
The content of the paper: In the rst section we present our key denitions
introducing systems of local lters and corresponding quasi-orders Q

(T), Q
0

(T).
In 1.6, 1.8 we explain how those partial orders can be made (<
+
)complete (as
we will be interested in directed subfamilies of Q

(T) and/or Q
0

(T)). A directed
subfamily H of Q

(T) and/or Q
0

(T) determines a lter l(H) on . One of the


basic questions is: how does the choice of H and/or of the system of local lters T
inuence the properties of the lter l(H)? Does the choice of T matter?
In Theorem 1.9 we show that ultralters generated by suciently directed gen-
erating systems are weakly reasonable, unless they are produced from a measurable
ultralter. This result can be used as an argument that our re-interpretation of the
Ppointness is very natural for weakly reasonable ultralters.
The second section is concerned with the full system T
ult
of local ultralters
and the ultralters on generated by H Q

(T
ult
). We show that there may be
weakly reasonable ultralters on generated by some H

Q
0

(T) which cannot


be obtained by the use of T
ult
(see Theorem 2.3). Thus, in particular, the choice
of the system of local lters may be important. Then we introduce more properties
of families H Q

(T
ult
) which are useful in generating ultralters on .
In the third section we introduce pararegular lters (Denition 3.1) and the full
system of local pararegular lters T
pr
. It occurs that lters l(r) for r Q
0

(T
pr
)
are related to generating numbers (in standard sense) of lters on (see 3.6, 3.8).
The referee of Shelah [8] requested that importance of the inaccessibility of
in the assumptions of [8, Prop. 1.6(1)] is claried. We pay this debt in the last
section of the present paper and we show that one does need the assumption that
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


GENERATING ULTRAFILTERS IN A REASONABLE WAY 3
is inaccessible for that result: Theorem 4.8 shows that consistently there is a very
reasonable ultralter D on
1
such that Odd has a winning strategy in
D
.
Notation: Our notation is rather standard and compatible with that of classical
textbooks (like Jech [1]).
(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 an uncountable regular
cardinal.
(2) For two sequences , we write whenever is a proper initial segment
of , and when either or = . The length of a sequence is
denoted by lh().
(3) We will use letters D, E, F and d (with possible indexes) to denote lters
on various sets. Typically, D will be a lter on (possibly an ultralter),
while E, F will stand for lters on smaller sets. Also, in most cases d will
be an ultralter on a set of size less than .
For a lter F of subsets of a set A, the family of all Fpositive subsets of
A is called F
+
. (So B F
+
if and only if B A and B C ,= for all
C F.)
(4) In forcing we keep the older convention that a stronger condition is the
larger one. For a forcing notion P,
P
stands for the canonical Pname for
the generic lter in P. With this one exception, all Pnames for objects in
the extension via P will be denoted with a tilde below (e.g.,

, X

).
1. Generating a filter from systems of local filters
Here we present the general scheme of generating a lter on a regular uncount-
able cardinal by using smaller lters. Our approach is slightly dierent from the
one in [8, 2] and/or [4, 1], but the dierence is notational only (see 1.3 below).
Denition 1.1. (1) A system of local lters on is a family T such that
all members of T are triples (, Z, F) such that Z , [Z[ < ,
= min(Z) and F is a proper lter on Z,
the set
_
< :
_
Z, F
__
(, Z, F) T
__
is unbounded in .
If above for every (, Z, F) T, the set Z is innite and F is a non-principal
ultralter on Z, then we say that T is a system of local non-principal
ultralters.
(2) More generally, if is a property of lters, then a system of local lters
on is a system of local lters T such that for every (, Z, F) T, the
lter F has the property . The full system of local lters is the family
of all triples (, Z, F) such that < , Z , [Z[ < and F is
a proper lter on Z with the property (assuming that it forms a system
of local lters). The full system of local non-principal ultralters on is
denoted by T
ult

or just T
ult
(if is understood).
The next denition introduces the lters generated by some families of local
lters. As we have said in the introduction, our motivations have roots in forcings
with norms and this suggested us to use sometimes a forcing-like notation (e,g, Q

)
similar to that of [3]. It is also worth noticing that some families of generators may
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


4 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
be used as forcing notions - for instance (Q
0

) is the forcing used in the end of


[4, Sec. 1].
Denition 1.2. Let T be a system of local lters on .
(1) We let Q

(T) be the family of all sets r T such that


_
<
__
[(, Z, F) r : = [ <
_
and [r[ = .
For r Q

(T) we dene
l(r) =
_
A :
_
<
__
(, Z, F) r
__
A Z F
__
,
and we dene a binary relation

F
on Q

(T) by
r
1

F
r
2
if and only if (r
1
, r
2
Q

(T) and) l(r


1
) l(r
2
).
(2) We say that an r Q

(T) is strongly disjoint if and only if

_
<
__
[(, Z, F) r : = [ < 2
_
, and

_
(
1
, Z
1
, F
1
), (
2
, Z
2
, F
2
) r
__

1
<
2
Z
1

2
_
.
We let Q
0

(T) be the collection of all strongly disjoint elements of Q

(T).
(3) We write Q

, Q
0

for Q

(T
ult
), Q
0

(T
ult
), respectively (where, remember,
T
ult
is the full system of local non-principal ultralters).
(4) For a set H Q

(T) we let l(H) =


_
l(r) : r H.
Remark 1.3. (1) Note that if r Q
0

then there is r

Q
0

such that l(r

) =
l(r) and for some club C of we have
_
(, Z) :
_
d
__
(, Z, d) r

__
=
_
(, [, )) : C & = min
_
C ( + 1)
__
.
Thus Q
0

is essentially the same as the one dened in [8, Def. 2.5].


(2) If H Q

(T) is

directed, then D = l(H) is a lter on extending


the lter of co-bounded sets. We may say the that the lter D is generated
by H or that H is the generating system for D.
Denition 1.4. Suppose that
(a) X is a non-empty set and F is a lter on X,
(b) F
x
is a lter on a set Z
x
(for x X).
We let
F

xX
F
x
=
_
A
_
xX
Z
x
: x X : Z
x
A F
x
F
_
.
(Clearly,
F

xX
F
x
is a lter on

xX
Z
x
.) If X is a linearly ordered set (e.g. it is a set
of ordinals) with no maximal element and F is the lter of all co-bounded subsets
of X, then we will write

xX
F
x
instead of
F

xX
F
x
.
Proposition 1.5 (Cf. [8, Prop. 2.9]). (1) Let T be a system of local lters on
and p, q Q

(T). Then p

q if and only if there is < such that


_
(, Z, F) q
__
A F
+
__
>
_
(

, Z

, F

) p
__
A Z

(F

)
+
__
.
(2) Let p, q Q

. Then the following are equivalent:


(a) p

q,
(b) there is < such that
_
(, Z, d) q
__
A d
__
>
_
(

, Z

, d

) p
__
A Z

__
,
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


GENERATING ULTRAFILTERS IN A REASONABLE WAY 5
(c) there is < such that if (, Z, d) q, , and X =
_
(, Z

, d

)
p : Z

Z ,=
_
, then X ,= and there is an ultralter e on X such
that
d =
_
A Z : A
e

: (, Z

)((, Z

, d

) X)
_
.
The quasi-orders (Q

) and (Q
0

) are (<
+
)complete (cf. [8, Prop.
2.3(3)]). Moreover, by essentially the same argument we may show the following
observation.
Proposition 1.6. Assume that T is a system of local lters on such that
()
sum
F
if < is an innite cardinal and a sequence (

, Z

, F

) : < ) T
satises
( < < )(Z

),
then for some uniform lter F on we have
_

0
,

<
Z

,
F

<
F

_
T.
Then both Q

(T) and Q
0

(T) are (<


+
)complete (with respect to

).
It is worth noticing that in general Q

(T) and/or Q
0

(T) do not have to be


even complete. For instance, consider the full system of co-bounded lters T
0
;
it consists of all triples (, Z, F) such that Z , [Z[ < , sup(Z) / Z and
F is the lter of all co-bounded subsets of Z. Let C consist of all ordinals <
divisible by , and for C and m < let Z

m
= [ + m , + m + )
and F

m
be the lter of co-bounded subsets of Z

m
. For n < put
p
n
=
_
( +m , Z

m
, F

m
) : C & 0 < m < & 2
n
[m
_
.
Clearly p
n
Q
0

(T
0
) and p
n

p
n+1
for all n < . One may easily verify that the
sequence p
n
: n < ) has no

upper bound in Q

(T
0
).
There is a natural procedure which for a given system T of local lters on
generates a system T

T satisfying the condition ()


sum
F
of 1.6 (so then Q

(T

)
and Q
0

(T

) are suitably complete).


Denition 1.7. Assume that
(a) T is a system of local lters on ,
(b)

E = E

: is a cardinal &
0
< ), where each E

is a uniform lter
on .
We dene:
(1) An (

E, T)block is a pair (T,

D) such that
T
<
is a wellfounded tree,
if T max(T), then < :

) T = for some innite


cardinal < ,


D = (

, Z

, F

) : max(T)) T,
if , max(T) and <
lex
, then Z

(where <
lex
is the
lexicographic order of T).
(2) By induction on the rank of the tree T, for an (

E, T)block (T,

D) we
dene a lter

D(T) on

: max(T) (where

D = (

, Z

, F

) :
max(T))).
If rk(T) = 0, i.e., T = ) then

D(T) = F

.
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


6 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
Suppose rk(T) > 0. Let = < : ) T (so
0
< is a
cardinal). For < we put
T

=
<
: )

T and

D

= (

, Z

, F

) : max(T) & (0) = ).


Plainly, each (T

,

D

) is an (

E, T)block (and rk(T

) < rk(T)). We
dene

D(T) =
E

<

(T

).
(3) The

Eclosure of T is the family of all triples (, Z, D) such that <
and for some (

E, T)block (T,

D) we have
Z =
_
Z

: max(T) and D =

D(T) and = min(Z)
(where

D = (

, Z

, F

) : max(T))).
Proposition 1.8. Assume that
(a) T is a system of local lters on ,
(b)

E = E

:
0
< & is a cardinal ), where each E

is a uniform
lter on .
Then the

Eclosure of T is a system of local lters extending T and satisfying the
condition ()
sum
F
of 1.6.
Suppose that a system T

of local lters on includes all triples (, , d),


where < and d is the principal ultralter on . For a set A let p
A
=
(, , d) T

: A Q
0

(T

). Note that p
A
p
B
= p
AB
, so easily if D
is a lter on extending the co-bounded lter, then H
D
def
= p
A
: A D is a

directed family and l(H


D
) = D. If D is a normal lter on , then H
D
will
be also (<
+
)directed (with respect to

). Consequently, if is a measurable
cardinal, then we may nd a system T of local lters on and a (<
+
)directed
family H Q
0

(T) such that l(H) is an ultralter including all club subsets of


(so l(H) is not weakly reasonable). However, to have a quite directed family H
such that l(H) is a non-reasonable ultralter we do need a measurable cardinal.
Theorem 1.9. Suppose that T is a system of local lters on , and H
Q

(T) is a (<)directed family such that l(H) is an ultralter. If l(H) is not


weakly reasonable, then for some club C

of the quotient ultralter l(H)/C

is
(<)complete and it contains all clubs of .
Proof. Assume that the family H Q

(T) is (<)directed and l(H) is an ul-


tralter which is not weakly reasonable. Let

=

: < ) be an increasing
continuous sequence of ordinals below such that
0
= 0 and for every club C
we have that
_
[

,
+1
) : C
_
l(H). Now, for a club C of and p H put
S(p, C) =
_
C :
_
(, Z, F) p
__
[

,
+1
) Z F
+
__
.
Claim 1.9.1. For every club C and p H, the set S(p, C) is stationary.
Proof of the Claim. Assume towards contradiction that S(p, C) is non-stationary.
So we may choose a club C

C of such that
()
1
_
C

__
(, Z, F) p
__
Z [

,
+1
) F
_
.
Pick a club C

such that
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


GENERATING ULTRAFILTERS IN A REASONABLE WAY 7
()
2
_
(, Z, F) p
__
C

__
<

_
.
By the choice of

we know that

,
+1
) : C

l(H), so necessarily

,
+1
) : C


_
l(p)
_
+
. Thus we may pick (, Z, F) p and C

such that Z [

,
+1
) F
+
(remember ()
2
), contradicting ()
1
.
Claim 1.9.2. (1) If p

q, p, q H and C

C are clubs of , then [S(q, C

)
S(p, C)[ < .
(2) If A , then there are p H and a club C such that either S(p, C)
A or S(p, C) A.
Proof of the Claim. (1) Pick < so that
_
(, Z, F) q
__
A F
+
__
> ((

, Z

, F

) p)(A Z

(F

)
+
)
_
(remember 1.5) and let

< be such that <

and
_
(, Z, F) q
__

Z

_
. Suppose that S(q, C

. Then C

C and there is
(, Z, F) q such that [

,
+1
) Z F
+
. Since

, we also have >


and hence there is (

, Z

, F

) p such that [

,
+1
) Z Z

(F

)
+
. Hence we
may conclude that S(p, C).
(2) Assume A . Let A

=
_
[

,
+1
) : A
_
. Since l(H) is an ultralter,
then either A

or A

belongs to it. Suppose A

l(p) for some p H. Pick a


club C such that
() if (, Z, F) p and (sup(Z) + 1) C ,= , then A

Z F.
Suppose S(p, C), so C and for some (, Z, F) p we have [

,
+1
) Z
F
+
. It follows from () that A

Z F and therefore A. Thus S(p, C) A.


If A

l(H), then we proceed in an analogous manner.


Let
D =
_
A : [S(p, C) A[ < for some p H and a club C
_
.
It follows from 1.9.1 that all members of D are stationary and since H is directed
we may use 1.9.2(1) to argue that D is a lter on . By 1.9.2(2) we see that D
is an ultralter on (so it also contains all clubs as its members are stationary).
Since H is (<)directed and the intersection of < many clubs is a club, we may
conclude from 1.9.2(1) that D is a (<)complete ultralter.
Let C

: < (so it is a club of ). To complete the proof of the theorem


we are going to show that D = l(H)/C

. Since we already know that D is an


ultralter, it is enough to show that S(p, C) l(H)/C

for every p H and a


club C . So let C be a club, p H and S

=
_
[

,
+1
) : S(p, C)
_
.
If S

l(H), then we are done, so assume that S

/ l(H). Since l(H) is an


ultralter and H is directed, we may nd q H such that p

q and S

l(q).
Let < be such that
_
(, Z, F) q
__
sup(Z) Z S

F
_
.
Since [S(q, C) S(p, C)[ < , we may pick S(q, C) S(p, C) such that > .
Then [

,
+1
) S

but also there is (, Z, F) q such that [

,
+1
) Z F
+
,
and thus also S

Z F
+
. However, sup(Z)

> , so Z S

F by the choice
of , a contradiction showing that S

l(H) as required.
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


8 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
2. Systems of local ultrafilters
In this section we are interested in the full system T
ult
of local ultralters on
and Q

, Q
0

. The rst question that one may ask is whether weakly reasonable
ultralters on generated by some H Q
0

(T) can be obtained by the use of Q


0

.
It occurs that it does matter which system of local lters we are using.
Denition 2.1. A lter F on a set Z is called an unultra lter, if for every A F
+
there is B A such that both B F
+
and A B F
+
. The full system of local
unultra lters on will be denoted by T
unu
. (Thus T
unu
consists of all triples
(, Z, F) such that , = Z , [Z[ < , = min(Z) and F is an unultra lter on
Z.)
Observation 2.2. (1) If F is an unultra lter on Z, A F
+
, then F +A
def
=
B Z : B (Z A) F is an unultra lter.
(2) Suppose that is a limit ordinal, Z

: < is a family of pairwise


disjoint non-empty sets, F

is a lter on Z

(for < ). Then

<
F

is
an unultra lter on

<
Z

. (Remember the convention declared in the last


sentence of Denition 1.4.)
Theorem 2.3. Assume
<
= and 2

=
+
. There exists a

increasing
sequence p

: <
+
) Q
0

(T
unu
) such that
(a) l
_
p

: <
+

_
is a weakly reasonable ultralter on , but
(b) there is no p Q
0

with l(p) l
_
p

: <
+

_
.
Proof. Fix enumerations
Y

: <
+
& is limit ) of all subsets of , and
r

: <
+
& is limit ) of Q
0

, and

: <
+
& is limit ) of all increasing continuous sequences of ordinals
below ,

: < ).
By induction on <
+
we choose p

Q
0

(T
unu
) so that the following conditions
are satised for every limit ordinal <
+
.
(o) For n < , the element p
n
Q
0

(T
unu
) is
_
(, Z

, F

) : < is limit, Z

= [, +) and F

is the lter of co-nite subsets of Z

_
.
(i) If cf() < , then for some increasing and conal in sequence
i
: i <
cf()), for every (, Z, F) p

, there is a sequence (
i
, Z
i
, F
i
) : i < cf())
such that
(
i
, Z
i
, F
i
) p
i
,
Z
i

j
for i < j < cf(),
Z =

i<cf()
Z
i
and F =

F
i
: i < cf().
(ii) If cf() = , then for some increasing and conal in sequence
i
: i < ),
if (, Z, F) p

and otp
__

< :
_
Z

, F

__
(

, Z

, F

) p

___
= j, then
(, Z, F) p
j
and
_
i < j
__
A F
+
__
(, W, D) p
i
__
A W D
+
_
.
(iii) If [(, Z, F) p

: Y

Z F
+
[ = , then
p
+1
=
__
, Z, F + [Y

Z]
_
: (, Z, F) p

& Y

Z F
+
_
,
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


GENERATING ULTRAFILTERS IN A REASONABLE WAY 9
and otherwise p
+1
=
_
(, Z, F) p

: Z Y

F
_
.
(iv) p
+2
p
+1
and for some club C of , for every C we have
Z

whenever (, Z, F) p
+2
, <

, and

+1
< min
_

:
_
Z, F
__
(, Z, F) p
+2
__
.
(v) p
+3
=
_
(, Z, F +A

) : (, Z, F) p
+2
_
where for every (, Z, F) p
+2
the set A

F
+
is such that
_
(, Y, d) r

__
A

Y / d
_
.
(vi) p
+3+n
= p
+3
for all n < .
(vii) For every r Q
0

and <
+
, if (, Z, F) p

and A F
+
, then there is
A

A such that
A

F
+
and
_
(, Y, d) r
__
A

Y / d
_
.
Conditions (o)(vi) fully describe how the construction is carried out and 1.5+2.2
imply that p

: <
+
) Q
0

(T
unu
) is

increasing. However, we have to argue


that the demand in (vii) is satised, as it is crucial for the possibility of satisfying
the demand in (v). Let r Q
0

. By induction on < we show that for every


(, Z, F) p

we have
()
(,Z,F)
if A F
+
, then there is A

A such that A

F
+
and
_
(, Y, d)
r
__
A

Y / d
_
.
(For a set A

as above we will say that it works for F and r.)


Step < .
Note that for each limit ordinal < there is at most one (, Y, d) r such that
Y [, + ) is innite. Assume A [, + ) is innite. Considering any two
disjoint innite sets A

, A

A we easily see that one of them must work for the


lter of co-nite subsets of [, +) and r.
Step = +n + 1, <
+
is limit, n < .
If (, Z, F) p
+n
, A F
+
, A A

and A

A works for F and r, then also A

works for F +A

and r.
Step = <
+
is limit.
Suppose that (, Z, F) p

. If cf() = , then (, Z, F) p

for some

< (see
(ii)) so the inductive hypothesis applies directly. So assume that cf() < . Then
Z =

i<cf()
Z
i
and F =

F
i
: i < cf() for some sequence (
i
, Z
i
, F
i
) : i < cf())
such that
()
(i,Zi,Fi)
holds for each i < cf(), and
Z
i

j
for i < j < cf().
Let

= sup(Z) and let A F


+
. Now we consider three cases.
Case A: For some

<

we have
_
(, Y, d) r
__
Y [

) =
_
.
Plainly, the set A

= A

works for F and r.


Case B: For some (, Y, d) r we have <

sup(Y ).
For each i < cf() such that A Z
i
(F
i
)
+
choose disjoint sets A
0
i
, A
1
i
(F
i
)
+
included in A Z
i
(remember each F
i
is an unultra lter) and let
A

=
_
A

i
: i < cf() & A Z
i
(F
i
)
+
A
(for < 2). Both A
0
F
+
and A
1
F
+
, and one of these two sets works for F
and r.
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


10 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
Case C: For each

<

there is (, Y, d) r such that

< < sup(Y ) <

.
Let A F
+
. Then the set I = i < cf() : A Z
i
(F
i
)
+
is unbounded in
cf() and using the assumptions of the current case we may choose an increasing
sequence i
j
: j < cf()) I such that for every (, Y, d) r there is at most one
j < cf() such that Z
ij
Y ,= . For each j < cf() pick A

ij
(F
ij
)
+
included in
A Z
ij
which works for F
ij
and r, and then put A

j<cf()
A

ij
.
Problem 2.4. Is it provable in ZFC that for some system T of local lters on
there exists a

directed family H Q

(T) such that


(a) l(H) is a weakly reasonable ultralter on , but
(b) there is no

directed family H

Q
0

(T
ult
) such that l(H) = l(H

)?
The assumption that a generating system H Q

(T) is directed is an easy way


to ensure that l(H) is a lter on . However, if we work with H Q

we may
consider alternative ways of guaranteeing this.
Denition 2.5. For p Q

let
(p) =
_
(, Z, d) T
ult
:
_
A d
__
(

, Z

, d

) p
__
A Z

__
.
Observation 2.6. (1) If p, q Q

, then p

q if and only if [q(p)[ < .


(2) If p Q

and (, Z, d) (p), then for some (


x
, Z
x
, d
x
) : x X p
and an ultralter e on X we have
d =
_
A Z : A
_
xX
Z
x

e

d
x
: x X
_
.
Denition 2.7. We say that a non-empty family H Q

is
(a) big if for each T T
ult
there is q H such that either q T or q T = ;
(b) linked if for each p
0
, . . . , p
n
H, n < , we have
_
:
_
Z, d
__
(, Z, d) (p
0
) . . . (p
n
)
__
[ = .
The property introduced in Denition 2.7(a) resembles the bigness of creating
pairs (see [3, Sec. 2.2]), so the use of the term big seemed natural. The name linked
is motivated by Observation 2.8(1) below.
Observation 2.8. (1) If H Q

is linked, then
(a) for each p
0
, . . . , p
n
H, n < , there is q Q

which is

above all
p
0
, . . . , p
n
,
(b) l(H) has nite intersection property.
(2) If H Q

is linked and big, then l(H) is an ultralter on .


For basic information on the ideal M

,
of meager subsets of

and its covering
number we refer the reader e.g. to Matet, Roslanowski and Shelah [2, 4]. Let us
recall the following denition.
Denition 2.9. (1) The space

is endowed with the topology obtained by
taking as basic open sets and O
s
for s
<
, where O
s
= f

: s
f.
(2) The (<
+
)complete ideal of subsets of

generated by nowhere dense
subsets of

is denoted by M

,
.
(3) cov(M

,
) is the minimal size of a family / M

,
such that

/ =

.
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


GENERATING ULTRAFILTERS IN A REASONABLE WAY 11
Theorem 2.10. Assume that =
<

1
and cov(M

,
) = 2

. Then there
exists a linked and big family H Q
0

such that l(H) is a weakly reasonable


ultralter.
Proof. The proof is very similar to that of [8, Thm 2.14]. Let be a suciently
large regular cardinal and let N H() be such that [N[ = and
<
N N.
Put T
ult
N
= T
ult
N. We will inductively construct a linked and big family H
included in Q
0

(T
ult
N
) Q
0

(T
ult
). The following two claims are the key points of
the inductive process. Below, linked means linked as a subfamily of Q

(i.e.,
it is the notion introduced in Denition 2.7(b)).
Claim 2.10.1. Assume that H
0
Q

(T
ult
N
) is linked, [H
0
[ < cov(M

,
), and
T T
ult
. Then there is q Q
0

(T
ult
N
) Q
0

such that H
0
q is linked and either
q T or q T = .
Proof of the Claim. We consider two cases.
Case A: For every n < , p
0
, . . . , p
n
H
0
and < there is (, Z, d) (p
0
)
. . . (p
n
) T T
ult
N
such that < .
Let T
0
be the family of all sequences such that
(i) lh() < ,
(ii) if < lh(), then () T T
ult
N
,
(iii) if <

< lh(), () = (, Z, d), (

) = (

, Z

, d

), then Z

.
It follows from the assumptions of the current case that T
0
is a branching tree
(remember [T
ult
N
[ = ). Moreover, for each p
0
, . . . , p
n
H
0
we have
_
lim(T
0
) :
_
<
__
>
__
() / (p
0
) . . . (p
n
)
__
M

,
.
Hence (as [H
0
[ < cov(M

,
)) we may pick lim(T
0
) such that for every p
0
, . . . , p
n

H
0
, n < , we have
[
_
< : () (p
0
) . . . (p
n
)
_
[ = .
Let q = () : < . Then q Q
0

(T
ult
N
) Q
0

, H
0
q is linked and q T.
Case B: Not Case A.
Then for some p

0
, . . . , p

m
H
0
and < we have
_
(, Z, d) (p

0
) . . . (p

m
) T
ult
N
__
> (, Z, d) / T
_
.
It follows from the choice of N that
if p
0
, . . . , p
n
Q

(T
ult
N
) and (, Z, d) (p
0
) . . . (p
n
),
then there are Z

, d

such that (, Z

, d

) (p
0
) . . . (p
n
) T
ult
N
.
Consequently, we may repeat arguments of the previous case replacing in clause
(ii) T T
ult
N
by T
ult
N
T. Then we obtain q Q
0

(T
ult
N
) Q
0

such that H
0
q
is linked and q T = .
Claim 2.10.2. Assume that H
0
Q

(T
ult
N
) Q

is linked, [H
0
[ < cov(M

,
) and
a sequence

: < ) is increasing continuously. Then there are p Q


0

(T
ult
N
)
and a club C

of such that
(a) H
0
p is linked, and
(b)
_
[
+1
,

) : < are successive members of C

_
l(p).
Proof of the Claim. This is essentially [8, Claim 2.14.4].
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


12 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
Now we employ a bookkeeping device to construct inductively a sequence q

:
< 2

) Q
0

(T
ult
N
) such that
for each < 2

the family q

: < is linked,
if T T
ult
N
, then for some < 2

we have q

T or q

T = ,
if

: < ) is increasing continuous, then for some < 2

and a
club C

of we have that
_
_
[
+1
,

) : < are successive members of C

_
l(q

).
Since [T
ult
N
[ = , so there are no problems with carrying out the construction.
It should be clear that at the end the family q

: < 2

is linked, big and it


generates a weakly reasonable ultralter.
Note that we may modify the construction in the proof of Theorem 2.10 so
that the resulting H is directed. Namely, by an argument similar to the one in
Claim 2.10.1 we may show, that if H
0
Q

(T
ult
N
) is linked, [H
0
[ < cov(M

,
)
and p
0
, p
1
H
0
, then there is q Q
0

(T
ult
N
) such that q (p
0
) (p
1
) and
H
0
q is linked. With this claim in hands we may modify the inductive choice
of q

: < 2

) so that at the end q

: < 2

is directed. However, we do not


know how to guarantee the opposite, that the family q

: < 2

is not directed
or even better, that for no directed H Q
0

do we have l(H) = l(q

: < 2

).
Thus the following question remains open.
Problem 2.11. Does H Q
0

is linked and big imply that H is directed?


3. Systems of local pararegular filters
In this section we are interested in lters associated with the full system T
pr
of
local pararegular lters on and we show their relation to numbers of generators
(in standard sense) of some lters on .
Denition 3.1. Suppose that Z is an innite set, = min(Z). A pararegular
lter on Z is a lter F on Z such that for some system A
u
: u []
<
) of sets
from F we have:
[ +[ < , and if u v []
<
, then A
v
A
u
,
if U is innite, then

A
{}
: U = , and
T =
_
B Z :
_
u []
<
__
A
u
B
__
.
If the cardinal above satises 2
|+|
< , then we say that the lter F is
strongly pararegular.
The full system of local pararegular lters on will be denoted by T
pr
and the
full system of local strongly pararegular lters on is denoted by T
spr
. (The latter
forms a system of local lters if and only if is inaccessible and then T
spr
T
pr
.)
Let us recall the following strong
+
chain condition.
Denition 3.2 (See Shelah [5, Def. 1.1] and [7, Def. 7]). Let Q be a forcing
notion, and < be a limit ordinal.
(1) We dene a game
cc
,
(Q) of two players, Player I and Player II. A play
lasts steps, and at each stage < of the play sequences p

, q

and a
function

are chosen so that:


p
0
=
Q
: i <
+
),
0
:
+

+
: i 0;
If > 0, then Player I picks p

such that
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


GENERATING ULTRAFILTERS IN A REASONABLE WAY 13
(i) p

= p

i
: i <
+
) Q satises ( < )(i <
+
)(q

i
p

i
),
(ii)

:
+

+
is regressive, i.e., (i <
+
)(

(i) < 1 +i);


Player II answers choosing a sequence q

= q

i
: i <
+
) Q such
that (i <
+
)(p

i
q

i
).
If at some stage of the game Player I does not have any legal move, then he
loses. If the game lasted steps, Player I wins a play p

, q

: < )
if there is a club C of
+
such that for each distinct members i, j of C
satisfying cf(i) = cf(j) = and ( < )(

(i) =

(j)), the set q

i
: <
q

j
: < has an upper bound in Q.
(2) The forcing notion Q satises condition ()

if Player I has a winning


strategy in the game
cc
,
(Q).
Proposition 3.3 (See Shelah [5, Iteration Lemma 1.3] and [7, Thm 35]). Let <
be a limit ordinal, =
<
. Suppose that

Q = P

, Q


: < ) is a (<)support
iteration such that for each <


satises ()

.
Then P

satises ()

.
Denition 3.4. Suppose that D is a uniform lter on . We dene a forcing notion
Q
pr
D
by:
a condition is a tuple p =
_

p
,
p

:
p
), Z
p

, F
p

: <
p
), /
p
_
such that
() /
p
D, [/
p
[ < ,
p
< ,
()
p

:
p
) is an increasing continuous sequence of ordinals below ,
() Z
p

= [
p

,
p
+1
) and F
p

is a pararegular lter on Z
p

;
the order
Q
pr
D
= is given by: p q if and only if (p, q Q
pr
D
and)
(i) /
p
/
q
,
p

q
,
(ii)
q

=
p

for
p
, and Z
p

= Z
q

, F
q

= F
p

for <
p
,
(iii) if A /
p
and
p
<
q
, then A Z
q

F
q

.
Proposition 3.5. Assume
<
= and let D be a uniform lter on . Then:
(1) Q
pr
D
is a (<)complete forcing notion of size 2

,
(2) Q
pr
D
satises the condition ()

of 3.2 for each limit ordinal < ,


(3) if r

is a Q
pr
D
name such that

Q
pr
D
r

=
_
(
p

, Z
p

, F
p

) : <
p
& p
Q
pr
D
_
,
then
Q
pr
D
r

Q
0

(T
pr
) and D l(r

) .
Proof. (1) Note that if < < , then there are

<
2
|[,)|
many pararegular
lters on [, ). Hence easily [Q
pr
D
[ = 2

.
If p

: < ) Q
pr
D
is
Q
pr
D
increasing, < , then letting /
q
=

<
/
p
,

q
= sup(
p
: < ) and
q

, Z
q

, F
q

: <
q
) =
_
<

, Z
p

, F
p

: <
p
)
we get a condition q = (
q
,
q

:
q
), Z
q

, F
q

: <
q
), /
q
) Q
pr
D
stronger than
all p

(for < ).
(2) Let A consists of all sequences Z

, F

: < ) such that Z

, F

: < ) =
Z
p

, F
p

: <
p
) for some p Q
pr
D
. By what we said earlier, [A[ = , so we may
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


14 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
x an enumeration

: < ) of A. Now, let st be a strategy of Player I in

cc
,
(Q
pr
D
) which, at a stage < of the play, instructs her to choose a legal inning
p

such that if i <


+
, then Z
p

, F
p

: <
p

i
) =

(i)
. (Note that
there are legal innings for Player I by the completeness of the forcing proved in
(1) above.) Plainly, if p

, q

: < ) is a play of
cc
,
(Q
pr
D
) in which Player
I follows st and i < j <
+
are such that

(i) =

(j) for all < , then


the family p

i
, p

j
: < has an upper bound. Thus st is a winning strategy for
Player I.
(3) Suppose p Q
pr
D
and let = [/
p
[ + [ +
p

p[. Fix a sequence A

: < )
listing all members of /
p
(with possible repetitions) and let u

: < ) be
an enumeration of []
<
. By induction on < choose an increasing sequence

: < ) [
p

p, ) such that



u
A

. (Remember, D is a uniform lter


and < .) Let
q
=
p
+ 1,
q

q = sup(

: < ) + 1 and for u []


<
let
B
u
=

: u u

& < . Then F


q

p
= B [
q

p
,
q

q
) : (u []
<
)(B
u
B)
is a pararegular lter on [
q

p ,
q

q ) and A [
q

p,
q

q ) F
q

p for all A /
q
. So
now we may take a condition q Q
pr
D
stronger than p and such that Z
q

p = [
p
,
q
),
/
p
= /
q
. Then q (
q

p
, Z
q

p
, F
q

p
) r

.
So we easily conclude that indeed
Q
pr
D
r

Q
0

(T
pr
) and D l(r

) (remember
the denition of the order on Q
pr
D
, specically 3.4(iii)).
Corollary 3.6. Assume
<
= , 2

=
+
, 2

+
=
++
. Then there is a (<)
complete
+
cc forcing notion P such that

P
2

=
++
and if D is a uniform lter on generated by
less than
++
elements, then D l(r) for some r Q
0

(T
pr
) .
Proof. Using a standard bookkeeping argument build a <support iteration

Q =
P

, Q


: <
++
) such that
for each <
++
we have that
P


= Q
pr
D

for some P

name D

for
a uniform lter on ,
if A

: <
+
) is a sequence of P

++names for subsets of , then for


some <
++
such that every A

is a P

name we have

if A


: <
+
) generates a uniform lter D on , then Q


= Q
pr
D
.
Now look at the limit P

++ = lim(

Q) (and remember 3.5, 3.3).


Proposition 3.7. Assume 2

=
+
. Then there is a uniform ultralter D on
containing no l(p) for p Q

(T
pr
).
Proof. First note that if F is a pararegular lter on Z, then for each we have
Z F. Consequently, if / []

is a family with p, [, ) : < /,


[/[ , and p Q

(T
pr
), then we may choose A such that
/ A has p,
for each (, Z, F) p we have [Z A[ 1 so also Z A F.
Hence, by induction on <
+
, we may choose a sequence A

: <
+
) of
unbounded subsets of such that
for < , A

= [, ),
A

: <
+
has p,
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


GENERATING ULTRAFILTERS IN A REASONABLE WAY 15
for every A there is <
+
such that either A

A or A

A = ,
for every p Q

(T
pr
) there is <
+
such that A

l(p).
Then D = A : A
0
. . . A
n
A for some
0
, . . . ,
n
<
+
, n < is an
ultralter as required.
Proposition 3.8. Assume that
(a) there exists a Kurepa tree with 2

branches,
(b) D is a uniform lter on ,
(c) p Q
0

(T
pr
) is such that l(p) D,
(d) if is a limit cardinal, then it is strongly inaccessible and p Q
0

(T
spr
).
Then the lter D cannot be generated by less than 2

sets, i.e., for every family


A D of size less than 2

there is a set A D such that [X A[ = for all


X A.
Proof. Let T be a Kurepa tree with 2

branches (so each level in T is of size


< ). For < let T

be the
th
level of T. Choose an increasing continuous
sequence

: < ) such that if (, Z, F) p and

<
+1
, then
Z
+1
and
there is a system A

u
: u [

]
<
) of sets from F witnessing that F is
pararegular (strongly pararegular if is inaccessible) with

satisfying
[T

.
For each < and (, Z, F) p such that

<
+1
, let us x an injection

: T

11

, and next for every branch through T let us choose a set A

D
so that
if < , T

, (, Z, F) p,

<
+1
, then A

Z = A

()}
.
For our conclusion, it is enough to show that if B D, then there are at most
nitely many branches through T such that [B A

[ < . So suppose towards


contradiction
0
,
1
,
2
, . . . are distinct branches through T, B D and [B
A
n
[ < for each n < . The set (, Z, F) p : B Z F
+
is of cardinality ,
so we may nd < and
n
T

(for n < ) such that



n
T

=
n
and
n
,=
m
for distinct n, m, and
B Z

(F

)
+
for some (

, Z

, F

) p satisfying

<
+1
, and
B

A
n
for all n < .
Then ,= B Z

(n)}
: n < , a contradiction.
4. Forcing a very reasonable ultrafilter
Our goal here is to show that the inaccessibility of in the assumptions of [8,
Prop. 1.6(1)] is needed. This answers the request of the referee of [8] and fullls
the promise stated in [8, Rem. 1.7]. Assuming that is strongly inaccessible, we
will construct a CS iteration P

, Q

: < ) of proper forcing notions such that

P
there is a (
1
)directed family H Q
0
1
such that
l(H) is a weakly reasonable ultralter on
1
and yet
Odd has a winning strategy in
l(H)
.
Let us recall the following denition.
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


16 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
Denition 4.1 (Shelah [8, Def. 1.4]). Let D be a uniform ultralter on . We
dene a game
D
between two players, Odd and Even, as follows. A play of
D
lasts
steps and during a play an increasing continuous sequence =
i
: i < )
is constructed. The terms of are chosen successively by the two players so that
Even chooses the
i
for even i (including limit stages i where she has no free choice)
and Odd chooses
i
for odd i. Even wins the play if and only if

[
2i+1
,
2i+2
) :
i < D.
The following result was shown in [8, Prop. 1.6]:
Proposition 4.2. Assume D is a uniform ultralter on .
(1) If is strongly inaccessible and Odd has a winning strategy in
D
, then D
is not weakly reasonable.
(2) If D is not weakly reasonable, then Odd has a winning strategy in the game

D
.
Before we dene our CS iteration P

, Q

: < ) let us introduce two main


ingredients used in the construction.
Sealing the branches: At each stage of the iteration we will rst use forcing
notions introduced in Shelah [6, Ch. XVII, 2].
For a tree T
<1

1
, the set of all
1
-branches through T will be denoted by
lim(T). Thus lim(T) =
1

1
: ( <
1
)( T).
Lemma 4.3 (Shelah [6, Ch. XVII, Fact 2.2]). Suppose that T
<1

1
is a tree
of height
1
. Let C be the Cohen forcing and L

be a Cname for the Levy collapse


of 2
2
to
1
(with countable conditions, so it is a closed forcing notion). Then

CL

lim(T) =
_
lim(T)
_
V
.
Denition 4.4 (Shelah [6, Ch. XVII, Def. 2.3]). Suppose that T
<1

1
is a
tree of height
1
, [T[ =
1
, [ lim(T)[
1
. Let B
i
: i <
1
) list all members of
lim(T) (with possible repetitions) and y
i
: i <
1
) list all elements of T so that
[y
j
y
i
j < i]. For j <
1
we dene
B

j
=
_
B
i
if j = 2i,
y
i
if j = 2i + 1,
and B

j
= B

j

_
i<j
B

i
.
Let w = j <
1
: B

j
,= and for j w let x
j
= min(B

j
). Finally, we put
A = x
i
: i w. We dene a forcing notion P
T
for sealing the branches of T:
a condition p in P
T
is a nite function from dom(p) A into such that if
, dom(p) and , then p() ,= p(),
the order
PT
of P
T
is the inclusion, i.e., p q if and only if (p, q P
T
and)
p q.
Lemma 4.5 (Shelah [6, Ch. XVII, Lem. 2.4]). Suppose that T
<1

1
is a tree
of height
1
, [T[ =
1
, [ lim(T)[
1
and P
T
is the forcing notion for sealing the
branches of T.
(a) P
T
satises the ccc.
(b) If G P
T
is generic over V and V

is a universe of ZFC extending V[G]


and such that (
1
)
V

=
V
1
(= (
1
)
V[G]
), then
V

[= lim(T) =
_
lim(T)
_
V
.
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


GENERATING ULTRAFILTERS IN A REASONABLE WAY 17
Adding a bound to ( Q
0
1
and a family | T(
1
): After sealing branches
of a tree, we will force a new member r

of our family H Q
0
1
at the same time
making sure that some family | of subsets of
1
is included in l(r

).
Denition 4.6. Suppose that ( Q
0
1
and | T(
1
) are such that
(a) ( Q
0
1
is

directed and
(b) U
0
. . . U
n

_
l(()
_
+
for every U
0
, . . . U
n
|, n < .
We dene a forcing notion Q
bd
((, |) as follows:
a condition p in Q
bd
((, |) is a triple (r
p
, G
p
, U
p
) such that r
p
T
ult
1
is countable
and strongly disjoint (i.e., it satises the demands of 1.2(2)), G
p
( is countable
and U
p
| is countable;
the order =
Q
bd
(G,U)
is dened by: p q if and only if (p, q Q
bd
((, |) and)
U
p
U
q
, G
p
G
q
, r
p
r
q
and for every (, Z, d) r
q
r
p
we have that:

_
(

, Z

, d

) r
p
__
Z


_
and

_
r G
p
__
(, Z, d) (r)
_
((r) was dened in Denition 2.5) and

_
U U
p
__
U Z d
_
.
We also dene a Q
bd
((, |)name r

by
Q
bd
(G,U)
r

=
_
r
p
: p
Q
bd
(G,U)
_
.
Lemma 4.7. Assume ( Q
0
1
, | T(
1
) satisfy demands (a),(b) of 4.6. Then
(1) Q
bd
((, |) is a closed forcing notion,
(2)
Q
bd
(G,U)
r

Q
0
1
and (r ()(r

) and | l(r

) .
Proof. (1) Straightforward.
(2) To argue that
Q
bd
(G,U)
r

Q
0
1
, suppose p Q
bd
((, |). Let r
n
:
n < = G
p
, U
n
: n < = U
p
(we allow repetitions). Choose inductively
(
m
, Z
m
, d
m
) T
ult
1
such that for m < we have

_
(

, Z

, d

) r
p
__
Z


0
_
, Z
m

m+1
, and
(
m
, Z
m
, d
m
) (r
0
) . . . (r
m
), and
U
0
. . . U
m
Z
m
d
m
.
[Why is the choice possible? Since ( is directed, we may rst choose s ( such
that r
0
, . . . , r
m

s. Then for some <


1
, if (, Z, d) s and , then
(, Z, d) (r
0
) . . . (r
m
). By the assumption 4.6(b) on | we know that
U
0
. . . U
m
Z d for
1
many (, Z, d) s, so we may choose (
m
, Z
m
, d
m
) s
as required.]
After the above construction is carried out, pick any uniform ultralter e on
and put
=
0
, Z =
_
m<
Z
m
, and d =
e

m<
d
m
.
Then q = (r
p
(, Z, d), G
p
, U
p
) Q
bd
((, |) is a condition stronger than p.
Thus by an easy density argument we see that
Q
bd
(G,U)
[r

[ =
1
. The rest
should be clear.
Let us recall that a very reasonable ultralter on is a weakly reasonable ultra-
lter D such that D = l(H) for some (<
+
)directed family H Q
0

(see [8, Def


2.5(5)]). Now we may state and prove our result.
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


18 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
Theorem 4.8. Assume that is a strongly inaccessible cardinal. Then there is a
cc proper forcing notion P such that

P
there is a

increasing sequence r

: <
2
) Q
0
1
such that
l
_
r

: <
2

_
is a very reasonable ultralter on
1
but Odd has a winning strategy in the game
{r

:<2}
.
Proof. The forcing notion P will be obtained as the limit of a CS iteration of proper
forcing notions P

, Q

: < ). The iteration will be built so that for each <

is a proper forcing notion of size < ,


so we will be sure that the intermediate stages P

and the limit P

will be proper
and each P

(for < ) will have a dense subset of cardinality < . Thus P

will
satisfy cc (and will not be collapsed). Since in the process of iteration we will
also collapse to
1
all uncountable cardinals below , we will know that

P

1
= (
1
)
V
& 2
1
=
2
= .
Thus we may set up a bookkeeping device that gives us a list C


, A

: < )
such that
C


is a P

name for a club of


1
,
A

is a P

name for a subset of


1
,

is a P

name for a function from


1
to
1
, and
for each P

name C

for a club of
1
, for some < we have
P
C

= C


,
and similarly for names A

for subsets of
1
and names

for elements of
1

1
.
Before continuing let us set some terminology used later. A partial strategy is a
function such that
dom()
<1

1
: lh() is an odd ordinal , and

_
dom()
__
()
1
(sup() + 1)
_
.
We say that a sequence
1

1
is played according to a partial strategy if
the sequence is increasing continuous, and
for every odd ordinal < lh() we have dom() and () = ().
If ,
1

1
and is played according to , then we say that = [] if (0) = (0)
and (2 + 2) = (2 + 1) + (1 + ) + 1 for each < . Also, for an increasing
sequence
1

1
let
U

=
_
__
(2), (2 + 1)
_
: <
1
_
.
Now, we will inductively choose Q

and T

, r

so that for each < the


following demands are satised.
()
1
r

is a P
+1
name for a member of Q
0
1
and
P
+1
( < )(r

),
()
2
T

is a P

name for a subtree of


<1

1
of height
1
(with no maximal
nodes).
()
3

is a P

name for a partial strategy with domain T

: lh() is odd ,
and all nodes of the tree T

are played according to

.
()
4

P
+1
_
lim(T

+1
)
__
=

+1
[

]
_
.
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


GENERATING ULTRAFILTERS IN A REASONABLE WAY 19
()
5

P

( < )(T

&

) and

P
+1
if
<1

1
is increasing continuous and such that
lh() = + 1 for a limit and ( < )( T

) but / T

,
then T

+1
.
()
6

P

0
, . . . ,
n
lim(T

)
__
<
__
U
0
. . . U
n

_
l(r

)
_
+
_
for each
n < , and

P
+1
_
lim(T

)
__
U

l(r

)
_
.
()
7

P
+1
A

l(r

) or
1
A

l(r

) and
P
+1
if

: < ) is
the increasing enumeration of C


, then for some club C


1
we have

1

_
[

,
+1
) : C

_
l(r

) .
()
8
For > 0, Q

is the P

name for the composition


C L

bd
(r

: < , U

: lim(T

))
(see 4.3, 4.4, 4.6). Hence we know that also
()
9
for every P
+1
name Q

for a proper forcing notion,


P
+1
Q

lim(T

) =
_
lim(T

)
_
V
P

.
To start, we let r
1
be any xed element of Q
0
1
. We choose

:
<1

1

1
so
that for every
<1

1
there is (, Z, d) r such that sup() < and Z

(),
and we let T

0
= T
0
=

[
0
] : <
1

<1

1
. (So T
0
is a tree with
lim(T
0
) =

[
0
].) Finally

0
=
0
=

T
0
: lh() is odd . Now, the
forcing notion Q
0
is:
C L

T0
Q

bd
(r
1
, U
0[0]
).
Clearly, the families r
1
and U
0[0]
satisfy the demands (a),(b) of Denition
4.6.
Now suppose that we have arrived to a successor stage = + 1 (and we have
already dened P

and P

names T

, and P
+1
names r

for < so that


the demands of ()
1
()
6
hold. It follows from ()
1
+ ()
6
that Q

is correctly
determined by clause ()
8
, so

= C L

bd
_
r

: < , U

: lim(T

)
_
.
(Remember also that, by ()
9
, all
1
branches of T

in extensions by proper forcing


over V
P

are the same as those in V


P

.) Note, that the last factor of Q

adds
an element r

Q
0
1
(see 4.7(2)) and we know that


_
<
__
r

_
and U

: lim(T

) l(r

) .
In V
P

, we may choose thin enough uncountable subset of r

, getting r

satisfying the demand in ()


7
and such that
_
(, Z, d), (

, Z

, d

) r

__
<

sup(Z) + <

_
.
Let

:
<1

1

1
be such that

dom(

) =

and for
<1

1
dom(

)
we have
()
1

() = min
_
<
1
:
_
(, Z, d) r

__
sup() < & Z
__
.
Let

] and let r

=
_
(, Z, d) r

: U

Z d
_
. It follows from our
choices so far that r

Q
0
1
, and r

for < , and also


8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


20 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
()
2
for each i <
1
,

(2i + 1) / T


_
(, Z, d) r

__

(2i) < & Z

(2i + 1)
_
.
Put T

= T

: <
1
and dene

:
<1

1

1
so that

and for
<1

1
T

we have
()
3

() = min
_
<
1
:
_
(, Z, d) r

__
sup() < & Z
__
.
Let
S

=
<1

1
: is increasing continuous of length lh() = + 1
for some limit and ( < )( T

) but / T

.
For each S

let

1
be such that

is played according to

and for every odd ordinal lh() we have

( + 1) =

() + 889. Put
T

+1
= T

: S

& <
1
. Note that (still in V
P

) we have that
lim
_
T

+1
_
= lim
_
T

: S

.
It follows from the choice of r

, r

that
_
lim(T

)
__
U

l(r

)
_
and by the
denition of

we get that
_
S

__
U

l(r

)
_
(remember the choice of r

).
Hence, remembering the denition of r

, we conclude that
_
lim(T
+1
)
__
U


l(r

)
_
.
Finally we put

+1
=

T
+1
: lh() is odd . One easily veries that
the relevant demands in ()
1
()
7
hold for T

+1
,

+1
and r

. Let us also stress


for future reference that
()
4
if dom(

+1
) T

is of length 2i + 1, then there is (, Z, d) r

such
that (2i) < and Z

+1
().
Suppose now that we have arrived to a limit stage < and we have dened
P

names Q

, T

, and r

for all < . In V


P

we dene T

<
T

and

<

. We have to argue that the relevant demands in ()


2
()
6
are satised,
and the only problematic one is the rst condition of ()
6
. If cf() =
0
, then

lim(T

) =

<
lim(T

), so there are no problems. We will show that ()


6
holds also if cf()
1
and for this we will argue a contrario.
Suppose towards contradiction that (cf()
1
and) we have P

names

0
, . . . ,

n
(n < ) and a condition p P

such that
p
P

0
, . . . ,

n
lim(T

) and
_
<
__

1
(U

0
. . . U

n
) l(r

)
_
.
Remembering that ()
1
+ ()
6
hold on earlier stages, we may pass to a stronger
condition (if necessary) and assume additionally that for some < , <
1
and
pairwise distinct
0
, . . . ,
n

1
we have
p
P

0
, . . . ,

n
/

<
lim(T

) and

0
=
0
, . . . ,

n
=
n
and
_
(, Z, d) r

__
sup(Z) U

0
. . . U

n
Z / d
_
.
The forcing notion P

is proper, so we may choose a countable elementary submodel


N H() such that

0
, . . . ,

n
,
0
, . . . ,
n
, , , , p, . . . N and then we may pick
an (N, P

)generic condition q p. Let

= N
1
and

= sup(N ), and we
may assume q P

. Then
()
5
q
P

(i n)( <

)( <

)(

i
/ T

) , and hence q
P

, . . . ,

/ T

.
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


GENERATING ULTRAFILTERS IN A REASONABLE WAY 21
Why? As for each N we have a name

N for an ordinal below


1
such
that p

/ T

, so we may use the genericity of q. By a similar argument,


()
6
q
P

(i n)( <

)( <

)(

i
T

) , so also q
P

( <

)(

0
, . . . ,

n
T

) ,
and
()
7
q
P

0
(

) = . . . =

n
(

) =

.
(Remember that

i
are increasing continuous.) Now, consider a P

name q

for the
following member of Q

:
_

C
,
L

,
PT

, (, r

, )
_
.
Directly from the denition of the order of the forcing Q
bd
and the choice of r

we see that
q (

, q

)
P

+1
r

(r

) .
It follows from ()
5
, ()
6
and ()
5
that
q (

, q

)
P

+1

0
(

+ 1), . . . ,

n
(

+ 1) T

+1
T

,
so let us look what are the respective values of the partial strategy

+1
. By ()
4
we know that
q (

, q

)
P

+1
there exists (A, Z, d) r

such that for each i n

i
(

) =

< and Z

i
(

+ 1) .
Since

> we get a contradiction with the choice of p.


This completes the inductive denition of the iteration and the names T

and r

. It should be clear that

P
the sequence r

: < ) is

increasing and
l
_
r

: <
2

_
is a very reasonable ultralter on
1
and

<

is a winning strategy for Odd in the game


{r

:<}
.

References
[1] Thomas Jech. Set theory. Springer Monographs in Mathematics. Springer-Verlag, Berlin, 2003.
The third millennium edition, revised and expanded.
[2] Pierre Matet, Andrzej Roslanowski, and Saharon Shelah. Conality of the nonstationary ideal.
Transactions of the American Mathematical Society, 357:48134837, 2005. math.LO/0210087.
[3] 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.
[4] Andrzej Roslanowski and Saharon Shelah. Reasonable ultralters, again. Notre Dame Journal
of Formal Logic, 52:113147, 2011. math.LO/0605067.
[5] Saharon Shelah. Strong Partition Relations Below the Power Set: Consistency, Was Sierpi nski
Right, II? In Proceedings of the Conference on Set Theory and its Applications in honor of
A.Hajnal and V.T.Sos, Budapest, 1/91, volume 60 of Colloquia Mathematica Societatis Janos
Bolyai. Sets, Graphs, and Numbers, pages 637668. 1991. math.LO/9201244.
[6] Saharon Shelah. Proper and improper forcing. Perspectives in Mathematical Logic. Springer,
1998.
[7] Saharon Shelah. Was Sierpi nski right? IV. Journal of Symbolic Logic, 65:10311054, 2000.
math.LO/9712282.
[8] Saharon Shelah. The combinatorics of reasonable ultralters. Fundamenta Mathematicae,
192:123, 2006. math.LO/0407498.
8
8
9


r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
9
-
1
7







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
9
-
1
7


22 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
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

You might also like