8

4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


SPECIALISING ARONSZAJN TREES AND PRESERVING SOME
WEAK DIAMONDS
HEIKE MILDENBERGER AND SAHARON SHELAH
Abstract. We show that (R, N, ) together with CH and all Aronszajn
trees are special is consistent relative to ZFC. The weak diamond for the
covering relation of Lebesgue null sets was the only weak diamond in the Cicho n
diagramme for relations whose consistency together with all Aronszajn trees
are special was not yet settled. Our forcing proof gives also new proofs to
the known consistencies of several other weak diamonds stemming from the
Cicho n diagramme together with all Aronszajn trees are special and CH.
The main part of our work is an application [15, Chapter V, 1 7] for a
special completeness system, such that we have a genericity game. Thus we
show new preservation properties of the known forcings.
1. Introduction
Let A and B be sets of reals and let E A B. Here we work only with Borel
sets A and B and absolute E, so that there are no diculties in the interpretation
of the notions in various ZFC models. The set A carries the topology inherited from
the reals and 2

carries the product topology. A function F : 2

<1
A is called
Borel function if each part F 2

, <
1
, is a Borel function. The complexity of
the set of
1
parts can be high.
Denition 1.1. (Denition 4.4. of [14]) Let (A, B, E) be the following statement:
For every Borel map F : 2
<1
A there is some g :
1
B such that for every
f :
1
2 the set

1
: F(f )Eg()
is stationary. Commonly, if E is not the equality (A, B, E) is called a weak dia-
mond.
The original diamond,
1
, is (A, B, E) with A = B = 2
<1
(so here we do
not have subsets of the reals), E being equality, in the special case of F being the
identity function. Jensen [9] showed that
1
holds in L. Devlin and Shelah [7]
showed that in the case [B[ = 2 some diamond principles follow from 2
0
< 2
1
.
In the mentioned work Jensen also showed that
1
implies the existence of a
Souslin tree. Since then it has been interesting to investigate which weakenings
of
1
still imply the existence of a Souslin tree. Moore, Hrus ak and Dzamonja
[14] introduce and investigate numerous versions of weak diamonds. Let Unif (/)
denote the relation (F

meager sets,

, ,), and let Unif (^) denote the relation

(G

null sets,

, ,). They show that (Unif(/)) implies the existence of a Souslin

tree, and from work by Hirschorn [8] they derive that (Unif(^)) does not imply
the existence of a Souslin tree. Another model (with larger continuum) is given by
Laver [11]. Since the Borel Galois-Tukey connections (see Vojt as [16]) in the Cicho n
Date: rev. of May 2008.
2000 Mathematics Subject Classication: 03E15, 03E17, 03E35.
The rst author was partially supported by the Austrian Fonds zur F orderung der wis-
senschaftlichen Forschung, grant no. 16334 and by travel support by the Landau Center.
The second authors research was partially supported by the United States-Israel Binational Sci-
ence Foundation (Grant no. 2002323). This is the second authors publication 848.
1
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


2 HEIKE MILDENBERGER AND SAHARON SHELAH
(Cov(^))

(Unif(/))

oo
(Cof(/))

oo
(Cof(^))
oo

, ,

oo
(Add(^)) (Add(/))
oo
(Cov(/))
oo
(Unif(^))
oo
Figure 1. The framed weak diamonds imply the existence of a
Souslin tree. The arrows indicate implications.
diagramme can be translated into implications of the corresponding weak diamonds
[14, Proposition 4.9], there is a Cicho ns diagramme of weak diamonds. So all its
entries above (Unif(/)) imply the existence of a Souslin tree, see Figure 1.
Also (

) together with all Aronszajn trees are special is consistent

relative to ZFC according to [12]. In this model, the continuum is
2
.
So, before this work, there was one question regarding the existence of Souslin
trees and the weak diamonds in Cicho ns diagramme left open: Does the weak
diamond for the covering relation (R, F

null sets, ) imply that there is a Souslin

tree? The answer is negative:
Theorem 1.2. (R, F

null sets, ) together with CH and with all Aronszajn trees

are special is consistent relative to ZFC.
Now we give an outline: An essential tool in the analysis of proper forcings are
countable elementary substructures: We let > 2
2
(this is the concrete inter-
pretation of the phrase suciently large in our context, and sometimes smaller
lower bounds suce, but let us be denite) be regular and denote by H() the set
of all sets of hereditary cardinality < . Let <

be a xed well-ordering of H()

such that x y implies x <

y. We work with countable elementary substructures

M (H(), ), and when we want to perform constructions along a well-order we
take M (H(), , <

). There are at most 2

0
isomorphism types of transitive
collapses (N, , (<

)
N
) of (M, , <

). By our proviso on <

, the relation (<

)
N
is still a well-order. In general we let the letter N (also with subscripts) stand for
transitive models (Mostowski collapses of the Ms), and let M stand for a countable
elementary submodel.
We shall dene a game played in countable parts of the iterated proper forcings
from [15, Chapter V, Section 5]. The countable elementary submodel M, P, p P
M, f

, . . . are parameters. The number of rounds is = otp(M ), where is the

iteration length. The generic player gives a real

and the antigeneric player gives

a real

dominating it in round < . The strategy of the game depends only on

the isomorphism type of the Mostowski collapse of the given countable elementary
submodel (M, P

, p), P

an iteration of length . In the central Theorem 3.4, we

prove the existence of a Borel functions B

: (

T() T() for <

1
,
such that B

has the play and the isomorphism type of the collapse as arguments
and then yields as value a bounded (M, P

)-generic lter i the generic player

wins. An (M, P

)-generic lter G is called bounded if there is a q P

such that
G = p M P

: p q. We will prove that there is a winning strategy for

the generic player and let the antigeneric player play in such a way that the generic
real or a Borel function applied to the generic real will be contained in the sets of
branches of a meagre measure zero tree. Then from
1
in V, which shows that
all the Mostowski collapses N and all used (nitely many) predicates on them are
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


SPECIALISING ARONSZAJN TREES AND PRESERVING SOME WEAK DIAMONDS 3
guessed stationarily often in
1
, we will derive that the extension preserves certain
weak diamonds. Juhasz question [13], whether (a denition can be found, e.g.,
in [15, Chapter I, Def. 7.1]) implies the existence of a Souslin tree, remains open.
It cannot be attacked by forcings adding no reals since in the presence of CH,
implies .
2. Proper forcings adding no new reals
We rst recall the denition of the forcings specialising an Aronszajn tree with-
out adding reals from [2] and [15, Chapter V, Section 6]. It is known that these
forcings are -proper for all <
1
and are D-complete for a simple
1
-completeness
system D, which guarantees that their countable support iterations do not add reals
[15, Theorem V.7.1]). Abraham gives a nice didactic exposition of the method of D-
completeness systems in [1, Section 5]. Here, we will take a simple
1
-completeness
system D similar to the one from Abraham and Shelahs work [2].
Jensen (see [6]) showed that the property of not adding reals is in general not
preserved in countable support iterations of proper forcings at limit steps of co-
nality . So some stronger requirement has to be imposed on the iterands. The
method of completeness systems that has been developed by Shelah [15, Chapter
V] is appropriate for our aim.
Recall, a specialisation of an Aronszajn tree T = (
1
, <
T
) is a function f :
1

Q such that for any s, t
1
, s <
T
t f(s) < f(t). We call such a function
monotone. Now we work with monotone functions f, that specialise only a part of
T, namely the union of countably many of its levels, so that the indices of the levels
form a closed set C. We call such a pair (f, C) an approximation. For <
1
let
T

denote the -th level of T. For x T

and < we let x be the y T

such
that y <
T
x. For making the notation easier, we consider only Aronszajn trees T
whose -th level, T

, is [, (+1)). This is no loss of generality since specialising

all these Aronszajn trees suces.
For any closed C of
1
, every monotone f :

C
T

Q can be extended to
a total specialisation (see, e.g., [8, Lemma 3.7]), and hence working with approxi-
mations on a closed set of levels is the same as working with all levels. We follow
the exposition in [2], where the promises (see Def. 2.3) are not only nite parts of
the Aronszajn trees as in the book [15], but they are functions from these nite
parts into Q. We follow the book [15] in that we use club sets of levels on which
the approximations will be dened and not just initial segments

as in [2].
We follow the Israeli convention that the stronger forcing condition is the larger
one. We assume that each poset P has a weakest element and denote it by 0
P
.
Denition 2.1. (A modication of [2, Denition 4.1].)
(1) An approximation is a pair (f, C) such that there is a countable ordinal
and C + 1, C is closed and C, f :

iC
T
i
Q is a partial
specialisation function. The ordinal is called last(f). We say (f
2
, C
2
)
extends (f
1
, C
1
) and write (f
1
, C
1
) (f
2
, C
2
) i f
1
f
2
and C
1
C
2
and
(C
2
C
1
) (

C
1
) = .
(2) We say H is a requirement of height <
1
i for some n = n(H) < , H
is a countable set of functions of the form h: dom(h) Q with dom(h)
[T

]
n
.
(3) We say that a nite function h: T

Q bounds an approximation f with

last(f) = i x dom(h), f(x) < h(x). More generally, if =
last(f), then h: T

Q bounds f i x dom(h)(f(x) < h(x)).

8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


4 HEIKE MILDENBERGER AND SAHARON SHELAH
(4) An approximation f with last(f) = is said to full the requirement H of
height i for every t [T

]
<
there is some h H which bounds f
and such that x : x dom(h) is disjoint from t.
If f fulls the requirement H, then any approximation f

with the same last

level that is dominated everywhere by f fulls the requirement as well. Note that
according to Denition 2.1(4) only innite requirements H can be fullled. For
= the necessary property is equivalent to having an innite set of pairwise
disjoint dom(h), h H and is equivalent to a property we call dispersedness:
Denition 2.2. H Q
[T]
n
is called dispersed i for each t [T

]
<
, there is
some h H such that t dom(h) = .
A forcing condition will be an approximation together with a T-promise. The
promises function as side-conditions and ensure that the forcing and also all of its
countable support iterations (see Theorem 2.20) do not add new reals.
In order to describe how elements of () are seen at lower levels in the tree,
we extend our -notation: Let < . For h: T

Q we let dom(h) T

and
h(x) = minh(y) : y = x, y dom(h). For a requirement H of height and
< we set H = h : h H.
Denition 2.3. (See [2, Denition 4.1 (4)].) is a T-promise i dom() is club
in
1
and = () : dom()) has the following properties:
(a) For each dom(), () is a countable set of requirements of height .
(b) ( dom())(H ()) H is dispersed.
(c) ( < dom())(() H : H ()). This condition implies
that H : ( > )(H ())) is countable.
Denition 2.4. ([2, Denition 4.1 (5)]) We say that an approximation (f, C) fulls
the promise i last(f) dom() and f fulls each requirement H in (last(f)).
Finally we can describe the iterands of our iteration of length
2
. Q
T
is called
o(T) in [2]. We do not know whether it is equivalent to the forcing notion Q
NNR
or NNR(T) from [15, V, 6.3]. NNR means no new reals.
Denition 2.5. ([2, 4.2]) Q
T
is the set of (f, C, ) such that (f, C) is an approx-
imation, and is a promise and (f, C) fulls . The partial order is dened as
(f
0
, C
0
,
0
) (f
1
, C
1
,
1
) i
(1) f
1
extends f
0
,
(2) C
1
is an end-extension of C
0
and C
1
C
0
dom(
0
), and
(3) ( dom(
0
last(f
1
))( dom(
1
) and
0
()
1
()).
If p = (f, C, ), we write f = f
p
, C = C
p
and =
p
, and we write last(p) =
last(f
p
) = max(C
p
).
Do not confound the countable, closed Cs that are the second coordinate of
the approximations with the true clubs dom() in
1
that are the domains of the
promises : the rst ones are approximations to the latter ones as in the forcing
adding a club through a stationary set by countable approximations [4]. However,
we take club sets dom() and not co-stationary sets as there, as we want to work
with proper forcings.
Now we want to extend a given condition to a stronger condition of a given
height, and we want to show that the set of promises can be enlarged.
Lemma 2.6. ([2, Lemma 4.3], The extension lemma.) Let <
1
. If p Q
T
and if last(p) < dom(
p
), then there is some q p such that
q
=
p
and
last(q) = . Moreover, if h: T

Q is nite and bounds f

p
, then q can be chosen
such that h bounds f
q
.
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


SPECIALISING ARONSZAJN TREES AND PRESERVING SOME WEAK DIAMONDS 5
Proof. The proof is done by induction on . First case: =
0
+ 1 is a successor.
We may assume that last(p) =
0
and we have to extend f
p
onto T
0+1
, fullling
all the countably many requirements in
p
(). We know that every requirement
H
0
for H
p
() is fullled by f
p
. So H
0
contains innitely many functions
h that bound f. We have countably many H, and we enumerate them as H
0
, H
1
,
. . . . There are enough points in T
0+1
T
0
such that in each H
i
there will be some
h
i
such that dom(h
i
)

dom(h
j
) : j < i = .
Since it will be used in the limit step, we now prove the moreover-clause. If h
bounds p as in the Lemma, we rst choose any extension p
1
of p with = last(p
1
)
and then we correct p
1
as follows to obtain q: There is some d > 0 such that
x dom(h), h(x) > f
p
(x
0
) + d. Now we take : Q
+
(0, d) order-preserving
and such that (x) < x for all x Q
+
. Now we set f
q
(x) = f
p
(x
0
) +(f
p1
(x)
f
p
(x
0
)). Hence h bounds q.
Second Case: is a limit of dom(
p
). We pick an increasing sequence of ordinals

i
, i < , converging to . We dene an increasing sequence p
i
Q
T
, i ,
beginning with p
0
= p and nite h
i
, g
i
: T

Q which bound p
i
and whose union of
domains will be T

. The passage from

i
to
i+1
uses the inductive assumption for

i+1
of the stronger claim in the moreover clause. The h
i
and g
i
ensure that f
q
is
bounded on each branch in T
<
and that f
q
on the level T

fulls all the promises

in (). Then we can dene q = (f, C, ) by C = C
p
and =
p
. We let
f

f
pi
: i < (x, limsup
i
f
pi
(x
i
)) : x T

. The values on level

might be irrational. We correct them to slightly larger values in Q that are so small
as to full all the promises in
q
() and let the resulting function be f
q
. Such a
choice is possible since all (, )-gaps in R are lled with sequences with values in
Q.
To carry out the step from i to i + 1, let
p
() = H
i
: i < . At step i,
we choose h
i
H
i
such that dom(h
i
)

dom(h
j
) : j < i = and we choose
g
i
g : [T

]
n(Hi)
Q : g(x) = f
p
(xlast(p)) +
1
2
i
and full both. In addition we
take care that

dom(h
i
) dom(g
i
) : i < = T

. Then we choose
i
so high that
dom(h
i

i
)

dom(h
j

i
) : j < i = . By the induction hypothesis of the
statement together with the moreover-clause we have some
i
> 0 and p
i
such
that for all j i, x dom(h
j
) f
pi
(x
i
) < h
j

i
(x
i
)
i
and last(p
i
) =
i
,
and the same can be arranged for the g
j
. Since h
j
H
j
is taken care of at each
step i j, in the end also f(x) < h
j
(x) for all x dom(h
j
).
Denition 2.7. Let p be a condition of height and let be a promise. We say
that p includes i dom() dom(
p
) and for all dom(), ()
p
().
If p includes , then p fulls . There is a sucient condition for the existence
of an extension q of p such that q includes :
Lemma 2.8. (Modication [2, Lemma 4.4.], Addition of promises.) Let p Q
T
and = last(p). Let be a promise with < = min(dom()) and dom()
dom(
p
). Suppose that for some nite g : T

Q called a basis for , g bounds f

p
and
( dom())(H ())(h H)(h = g).
Then there is an extension q of p in Q
T
that includes .
Proof. Since g is nite, there is some rational d > 0 such that (x dom(g))(g(x) >
f
p
(x) +d). Now every H () is a dispersed collection of functions h with h
g. Let p
1
be any extension of p of height . For we set
q
() = ()
p
(),
and [, ) we set
q
() = H : H ()
p
(). The desired exten-
sion of p is obtained by correcting f
p1
to get f
q
that fulls () () as in the
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


6 HEIKE MILDENBERGER AND SAHARON SHELAH
moreover-part of the previous lemma.
In the following lemma > 2
1
is suciently large.
Lemma 2.9. ([2], [15, Fact V 6.7]) Let T be an Aronszajn tree. Let M (H(), )
be a countable elementary substructure with a suciently large regular , Q
T
M,
p Q
T
M, =
1
M and h: T

Q be a nite function which bounds f

p
. Let
D M, D Q
T
be dense open. Then there is an q p, q DM, that h bounds
q.
Proof. We assume that the contrary is the case. Let T, M, p, h be a counterexam-
ple. Let
0
= last(p) < and let x
0
, . . . , x
n1
= dom(h) [T

]
n
. Let v
i
= x
i

0
.
We assume that v
i
,= v
j
for i ,= j otherwise we extend p upwards with Lemma 2.6
to get some p

p with last(p

) < and x
i
last(p

) ,= x
j
last(p

).
Put g
0
= h
0
. Then g
0
M, as it is nite. We say that that a nite partial
g : T

Q is bad i
0
and g
0
= g
0
and, whenever q D extends p and
last(q), g does not bound q. So g is bad i it has the similar behaviour as h
0
.
For every [
0
, ), h is bad. So in M and hence in H() there are uncountably
many bad gs. We set
B = dom(g) : g is bad.
Then B is an uncountable and closed downwards in <
T
(above
0
) subset of

0<1
[T

]
n
. As T is an Aronszajn tree, [6, Lemma VI.7] implies that there
is some
0
and some B
0
B such that:
(1) For
0
<
1
, B
0
T
0
= (B
0
T
1
)
0
(2) B
0
T

is dispersed.
Here we take X = x : x X for X T. We may nd B
0
in M, since only
parameters in M were mentioned in its denition. For <
1
let () = H

with H

= g : g is bad and dom(g) B

0
T

M
. By Lemma 2.8, read in M,
there is an extension q of p in M of height which includes , i.e., H

q
().
Now let r D be any condition extending q. Let = last(r). Since r fulls ,
for some g in H

, g bounds r. But this contradicts the fact that g is bad.

Lemma 2.9 will be used in the induction in Claim 2.16 to get point (5).
Denition 2.10. Now we assume V [= CH +
1
+ 2
1
=
2
and let P
2
=
P

, Q

:
2
, <
2
) be a countable support iteration with Q

= Q
T

being
as above for some Aronszajn tree T

V[G

], where the lter G

is P

-generic
over V, such that
P
T

is an Aronszajn tree and for <
1
its -th level is
[, + ). The book-keeping shall be arranged so that every P
2
-name for an
Aronszajn tree is used in some iterand.
Why does every Aronszajn tree in V
P
2
have a P

-name for some <

2
?
We have [Q
T
[ =
2
, so that we cannot work with the
2
-chain condition for each
iterand. Now [15, Chapter VIII, Section 2] helps: Basically by Lemma 2.8, each Q
T
has the
2
p.i.c. (proper isomorphism condition), see [15, Chapter VIII, Def. 2.1],
and hence by [15, Chapter VIII, Lemma 2.4], P
2
has the
2
-c.c, if V
0
fulls the
CH.
Since P
2
has the
2
-c.c., by a lemma similar to the one of [5, 5.10], now for
subsets of
1
instead of real numbers, every subset of
1
in a countable support
iteration of proper forcings with the
2
-c.c. at each initial segment has a name at
some stage of conality
1
. So we an carry out the desired book-keeping.
In the remainder of this section, we shall prove that P
2
does not add new
reals. Towards this aim, we rst recall some general theory for <
1
-proper forcings
P adding no reals. Then we shall show that our specic forcing and a suitable
completeness system D(M, P, p) exhibit these properties. Note that adding no reals
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


SPECIALISING ARONSZAJN TREES AND PRESERVING SOME WEAK DIAMONDS 7
and adding no new -sequence of ordinals is the same for proper forcings. In the
application, P is of the form Q
T
or is some countable support iteration of Q
T
s.
Recall, p P is (M, P)-generic if for every P-generic lter G over V with p G,
p M[G

] On = M On. Now in the context of proper forcings that do not add

reals we nd completely (M, P)-generic conditions.
Denition 2.11. A condition p is completely (M, P)-generic if G = q P M :
q p is an (M, P)-generic lter. G is called bounded.
Indeed, P is proper and does not add reals i for every M (H(), ), for every
p M P there is a completely (M, P)-generic q p. Given a name f

for a
real, consider the dense sets D
n
= p : (m )(p f

(n) = m). Completeness

systems that are closed under nite intersections we shall have countably closed
ones help to nd completely generic conditions in a rst order denable way
and allow to prove that no new reals sneak in at the limit steps. Only the case
of conality is hard, since every real in a countable support iteration of proper
forcings appears for the rst time at some stage of at most countable conality [1,
Cor. 2.9 (1)]. An important point is that some parameters of the members of the
completeness system, that are subsets of M, here called x, need to be guessed. Since
intersections over countable parts of the completeness system are not empty, the
guessing can be performed in M

, when M

(H(), , <

) and M M

. One not
so aesthetic feature stays: There is neither a completeness system for the two-step
iteration nor for the limit forcing, we only know that no reals are added. From the
proof we get a description of the bounded generic lters and of the generic lters
for some towers of elementary submodels that appear as helpers in the proofs.
Denition 2.12. ([15, V, 5.5])
(1) We call D a completeness system if for some , D is a function dened
on the set of triples M, P, p), p M P, P M, M (H(), ), M
countable, such that D(M, P, p) is a family of non-empty subsets of
Gen(M, P, p) =G : G M P, G is directed and p G
and G 1 , =
for every dense subset 1 of P which belongs to M.
(2) We call D a -completeness system if each family D(M, P, p) has the prop-
erty that the intersection of any i elements is non-empty for i < 1 + (so
for
0
, D(M, P, p) generates a lter).
1
-completeness systems are also
called countably closed completeness systems.
(3) We say D is on if M (H(), ). We do not always distinguish strictly
between D and its denition.
The notion of forcing Q
T
has size 2
1
, and the set of all approximations has size

0
1
. So for a countable M (H(), , <

), we never have P M. If T M, we
can read the denition of P = Q
T
in M and get P
M
. Since T is denable from Q
T
(x ,<
T
y i there is an approximation with f(x) = f(y)), Q
T
M implies T M.
If > 2
1
is regular, then P
M
= P M. In our description via rst order formulae,
P, x, and G are predicates on M.
Denition 2.13. Suppose that D is a completeness system on . We say P is D-
complete, if for every countable M (H(), ) with P M, D M, p P M,
the following set contains as a subset a member of D(M, P, p):
Gen
+
(M, P, p) = G Gen(M, P, p) : there is an upper bound for G in P.
Denition 2.14. ([15, V, 5.5])
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


8 HEIKE MILDENBERGER AND SAHARON SHELAH
(1) A completeness system D is called simple if there is a rst order formula
such that
D(M, P, p) = A
x
: x is a nitary relation on M, i.e., x M
k
for some k ,
where
A
x
= G Gen(M, P, p) : (M T(M), , p, M, P) [= (x, G).
(2) A completeness system D is called almost simple over V
0
(V
0
a class,
usually a subuniverse) if there is a rst order formula such that
D(M, P, p) = A
x,z
: x is a nitary relation on M, i.e.,
x M
k
for some k , z V
0
,
where
A
x,z
= G Gen(M, P, p) :
(V
0
M T(M),
V0
,
MPP(M)
, p, M, V
0
, P) [= (x, z, G),
where
A
= (x, y) A A : x y.
(3) If in (2) we omit z, we call D simple over V
0
.
We shall give an example and a simple
1
-completeness system D on any
regular > 2
2
, so that Q
T
is D-complete. From now on we use the requirement
from Def. 2.10 that the -th level of T = (
1
, <
T
) is [, ( + 1)). Let > 2
2
be a regular cardinal. If we have a countable M (H(), ), then M T = T
<
for = M
1
. We take an increasing sequence

=
n
: n ) that is conal
in . Now we take for x
1
M a code of the branches through T
<
, for example
x
1
: T
<
, x
1
is eventually constant on each branch. We also code in x
1
the
branches through T
<
that have <
T
successors in T

. Indeed the other branches

are unimportant. If we want to nd an (M, P)-generic condition with last level
T

we have to take care that the approximations to the specialisation function do

not diverge on any branch that is continued in T

. Since we are looking for a

condition q p and p M, we also code into another component x
2
M the set

q
() of promises for each q M P. The codes x = (x
1
, x
2
,

) are in
general not in M, but they are predicates M
k
. The point is that countably many
A
x
from Denition 2.14 (the appearing in A
x
will be given in Lemma 2.15) have
a non-empty intersection. This works also for countably many guesses for codes,
which is crucial in the proofs of Theorems 2.20 and 3.4.
The technique of the following lemma comes from [2]. Actually a sketch of the
elements of the
1
-completeness system is also given in the end of the proof of [15,
Chapter V, Theorem 6.1] on page 236. We conceive x = (x
1
, x
2
,

) as one relation
in M.
Lemma 2.15. Q
T
is D-complete for the simple
1
-completeness system D given by
(x, G) =
0
(x)
1
(x, G), with

0
(x) x = (x
1
, x
2
,

)

=
n
: n ) increasing
M
1
=
_

n
: n <
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


SPECIALISING ARONSZAJN TREES AND PRESERVING SOME WEAK DIAMONDS 9
and

1
(x, G) ( > 0)(m < )(n
1
< n
2
[m, ))(t T

)(y
1
, y
2
<
T
t)
_
(y
1
T
n
1
y
2
T
n
2
y
1
<
T
y
2
f

[G](y
2
) < f

[G](y
1
) +

2
n2
_
G is a lter
p G D M((D P D dense in P) D G ,= )
(H x
2
)(n)(t [T
n
]
<
)(h H)
(domh
n
t = f

[G] T
n
fulls h
n
).
Here M, P, x and G appear in the formulas as (names for) predicates and p is a
constant. To ease readability, we write T

instead of x
1
(though T

is not a subset
of M) and

p
() instead of x
2
.
Proof. First we proof the following claim:
Claim 2.16. Let = M
1
= sup
n
: n < ) and let the
n
be increasing. If
(M T(M),
MP(M)
, p, M, Q
T
) [=
0
(x),
then there is G Q
T
, G G(M, Q
T
, p) A
x
such that
(M T(M),
MP(M)
, p, M, Q
T
) [= (x, G).
Proof. Let I
n
: n be an enumeration of all open dense subsets of Q
T
that
are in M. Let t
n
: n enumerate T

: Now we choose by induction on n < ,

p
n
such that
(1) p
0
= p,
(2) p
n+1
p
n
M,
(3) last(p
n+1
)
n+1
,
(4) p
n+1
I
n
,
(5) (t t
k
: k n)(y <
T
t)
_
y T
n+1
f
pn+1
(y) < f
pn
(y
n
) +
1
2
n+1+n
_
.
Then G = r : (n )(r p
n
) Gen(M, Q
T
, p) A
x
.
Why is this choice possible? For Properties (4) and (5) we use Lemma 2.9 for h
with
dom(h) = t
k

n+1
: k n,
h(y) = f
pn
(y
n
) +
1
2
n+1+n
,
which is a nite function that bounds p
n
and we nd some p
n+1
of length
n+1
.
Claim 2.17. If (M T(M), , p, M, Q
T
) [= (x, G) for some x, then G has an
upper bound in Q
T
.
Proof. Again let I
n
: n be an enumeration of all open dense subsets of Q
T
that are in M. Let x be as in (x, G). Let G q
n
: n , q
n
M I
n
,
last(q
n
) =
n
such that the
n
and the q
n
are increasing. We set = M
1
=

n
, f
q
as in the proof of Lemma 2.6 a slightly larger rational variant of

f
qn

(z, supf
qn
(z
n
) : n ) : z T

, C
q
=

n
C
qn
, which is closed since
for each n, C
qn+1
is an end extension of C
qn
, dom(
q
) = (

n
dom
qn
[,
1
))
, and for

> ,
q
(

) =

n

qn
(

) and
q
() =

n

qn
(

).
We claim that q is an upper bound of G: First we check that q Q
T
. Note that
if dominates all h
,z
, z T

, then for every z T

the limit f
q
(z) exists, because
if h
z,

, then for almost all n, z

n
=
n
+h
z,

(n) and h
z,

(n) (n). So we
have that (f
q
, C
q
) is an approximation. Now let H
q
() be a T-promise. For
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


10 HEIKE MILDENBERGER AND SAHARON SHELAH
some

, k , H
q
k
(

). Then, since q
k
fulls the promise, also q fulls
the promise.
Proof of Lemma 2.15 continued: We showed that A
x
G
+
(N, Q
T
, p). So we
have that Q
T
is D-complete. It remains to show that D is countably closed, i.e.,
that given x

with (x

, G), < , the intersection

A
x

is not empty. But

this is now easy: Let x

= (x
1,
, x
2,
,

). x
1,
, coding the conal branches in T
<
,
and x
2,
, coding the promise (), are dened from T and p and do depend on at
most in the way the coding is chosen, not in the content they code.
There is only some little twist because the

=
,u
: u < ) are not the
same. We choose =
m
: m < ) such that
0
= 0, ( m)(u < )(
,u

[
m
,
m+1
)). Then we let x
1
= x
1,0
, x
2
= x
2,0
and x = (x
1
, x
2
,

). Then A
x
A
x

,
< .
Denition 2.18. We call P -proper if the following holds: Let M
i
, i < , be
countable elementary submodels of (H(), ). Let P M
0
and let M
i
: i < )
be an increasing sequence such that M
j
: j i) M
i+1
and for limit ordinals j,
M
j
=

i<j
M
i
. Then for every p P M
0
there is some q p that is (M
i
, P)-
generic for all i < . Such a sequence M
i
: i < ) is called a tower of models and
is the height or the length of the tower.
Lemma 2.19. Q
T
is -proper for all <
1
.
Proof. The upper bound from Claim 2.17 gives a completely (M, Q
T
)-generic q p.
Given a tower of height , we can repeat the construction steps, using a diag-
onalised version of Claim 2.16 for countably many M and countably many enu-
merations of dense sets simultaneously, so that in the end we get some q that is
(M
i
, Q
T
)-generic for all i < .
Now we can cite Theorem V.7.1 (2) of [15] for
1
-complete systems. A very clear
proof, even in a more general context when almost simple over V
0
is replaced by
in V
0
, is given in [1, Theorem 5.17].
Theorem 2.20. Let P

= P
j
, Q

i
: j , i < ) be a countable support iteration.
If each Q
i

is -proper for every <

1
and D
i
-complete for some almost simple

1
-completeness system D
i
over V
0
(not over the current stage of the iteration),
then P

does not add reals.

So we know that P
2
from Denition 2.10 exists and specialises all Aronszajn
trees and does not add reals. The remaining task is to obtain the weak diamond
(R, ^, ) in V
P
2
.
3. Games for the generic filters over countable models
In this section we show that certain weak diamonds hold when forcing with a
countable support iteration of Q
T
s (of arbitrary iteration length ) over a ground
model fullling
1
. In order to specialise all Aronszajn trees, we start with a
ground model of CH and 2
1
=
2
and perform an iteration of length =
2
with
a suitable book-keeping.
For the weak diamonds, we rework the facts used in the proof of Lemma 2.15
to give some stronger, descriptive statement about G M. The basic idea is:
The parameters x
1
and x
2
of the A
x
in the completeness system D(M, P, p) from
Lemma 2.16 can be coded into functions : in a way that each

also serves as a code for a parameter. The proof of Theorem 2.20, which works
with guessing parameters, will be translated into a game whose innings give

-
suciently large codes of parameters. Let be the iteration length. The result,
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


SPECIALISING ARONSZAJN TREES AND PRESERVING SOME WEAK DIAMONDS 11
stated in Theorem 3.4, is that bounded (M
0
, P

)-generic lters containing p

0
can
be computed in a Borel manner from the isomorphism type of (M
0
, P

, p
0
) and a
game played according to a strategy. The length of the game is = otp(M
0
).
In the following > 2
2
suces. Let <

be a xed well-ordering of H() such

that x y implies x <

y. Assume that M (H(), , <

) is a countable model
and T, Q
T
M. From now on we shall use the well-order <

. In the following, let

M always be a model of this kind. We reserve the letter N for transitive collapses of
the Ms. Fix a bijective pairing function e: that is so low in complexity
such that it is an element of every M.
Now we want to get rid of the two parameters x
1
and x
2
that depend on p, T

and

p
(

) and are relations over M but not elements in M. The trick is

to nd a real coding them (after a transitive collapse) and code in such a way
that every

codes even better. Coding means we want to imitate Lemma 2.16

now with taking the role of x
1
and of x
2
. The parameter

can stand as it is,
since it depends only on the transitive collapse of M and not on P and p.
We translate the task of x
1
:
Denition 3.1. Let T be an Aronszajn tree with levels T

= [, ( + 1)). Let
be a limit ordinal in
1
. Given

converging to , we can write conally many
nodes of a branch b of T
<
into a function h
b,

: , such that for all n,

b T
n
=
n
+h
b,

(n)
and we can describe each node t = + k T

, by h
t,

: , such that for all

n,
t
n
=
n
+h
t,

(n).
If t =
n
+k T
n
, then we dene h
t,

: n + 1 , such that for all m n,

t
m
=
m
+h
t,

(m).
Now we translate the task of x
2
:
Denition 3.2. Let = M
1
. Given

converging to , and p M Q
T
with
last(p) =
0
, let
p
() = H
n
: n , and let h
n,m
H
n
be such that p fulls
h
n,m
, and such that dom(h
n,m
) : m < is dispersed and pairwise disjoint. We
dene h
p,Hn
: T

, such that for all x T

, for all m
h
n,m
(x) 2
hp,Hn
(x)
> f
p
(xlast(p)).
That is, the growth of f
q
f
p
along the branch leading to x T

and a promise
H
n

p
() shall be bounded, only the small increase 2
hp,Hn
(x)
above f
p
is allowed.
We code the level T

M in a predicate on M and we code the promise

p
() into
the natural numbers via a bijection l : T

. Then h
p,Hn
l : is a function
we want to eventually dominate with a good parameter . The parameter does not
know the actual functions h
p,Hn
. That aim ist: if a parameter dominates all the
h
t,

, t T

, and all the h

p,Hn
, n . then we can choose the conditions in an
(M, P)-generic lter only with the knowledge of for any

the parameter and

without T

(or x
1
) and
p
() (or x
2
). To make the induction in the next lemma
going, the parameter need also to be larger than the codes of the
pn
() for n .
So we code all h
q,H
for H
q
(), q M P, into x
2
.
Lemma 3.3. Let p Q
T
M. Let = M
1
= sup
n
: n < ),
n+1
>
n
.
Let c: M be a bijection with c(0) = Q
T
, c(1) = p, c(2n + 2) =
n
, and let
U = U(M, Q
T
, p)
= 2e(n
1
, n
2
) : c(n
1
) c(n
2
) 2e(n
1
, n
2
) + 1 : c(n
1
) <

c(n
2
).
We let
p
() = H
n
: n and we let the functions h
y,

and h
p,Hn
be dened
as in Defs. 3.1 and 3.2.
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


12 HEIKE MILDENBERGER AND SAHARON SHELAH
There is a Borel function B
1
:

T() T(), such that for every

, if
(3.1) (y T

)(h
y,

),
and
(3.2) (n)(h
p,Hn
(l())

)
for
G = c(n) : n B
1
(, U)
the following holds: G is (M, Q
T
)-generic and p G and there is an upper bound
r of G as in Claim 2.17.
Remark: r is an upper bound of G i we have for every Q
T
-generic lter G
V
over V with r G
V
and name G
V

that
r
QT
G
V

M = c(n) : n B
1
(, U).
Proof. We verify that each step in the proof of Lemma 2.15 is Borel-computable
from (, U). Let M (H(), , <

) be countable. Then we take an enumeration

I
n
: n ) of all dense subsets of Q
T
that are in M, ordered according to <

.
Now, we compute from and U by induction on n < , p
n
such that
(1) p
0
= p, last(p) =
0
(2) p
n+1
is the <

-least element of M such that

(2a) p
n+1
p
n
,
(2b) last(p
n+1
)
n+1
,
(2c) p
n+1
I
n
,
(2d) (x T
n+1
)
_
h
x,

(n + 1) (n + 1) f
pn+1
(x) < f
pn
(x
n
) +
1
2
n+1+n+(l(x))
_
.
For nding such an p
n+1
we use the Lemma 2.9 for the nitely many initial segments
of branches y (
n+1
+1) with y(
n+1
) (n+1) and with the following bound h:
dom(h) = x T
n+1
: h
x,

(n + 1) (n + 1),
h(x) = f
pn
(x
n
) +
1
2
n+1+n+(l(x))
.
If Equations (3.1) and (3.2) hold, then is suciently large to take care of all
branches of T
<
that lead to points x T

. Set B
1
(, U) = q NQ
T
: (n)q
p
n
.
Then B
1
(, U) Gen
+
(N, Q
T
, p) A
x
and there is an upper bound of B
1
(, U)
as in Claim 2.17.
Strictly speaking we must write U = U(M, P, p,

), since by the boundedness
theorem (see, e.g., [10, Theorem 31.1]) a conal sequence

cannot be computed
in a Borel manner from (M, , <

), and for each n,

n
is coded by the stipulation
c(2n + 2) =
n
. The arguments (M, P, p) of U will change during the iteration,
and one of the main tasks is to show that all the changes are Borel computable,
see for example Equation (3.8). Fortunately, since in proper forcing P the ordinary
height of N and N[(] (we use the letters N and ( for the objects after the transitive
collapse) are the same for all (M, P)-generic lters G,

will not change and it does
not hide features of the proof if we do not write it during the proof of the iteration
theorem. However,

needs to be guessed as one component in Lemma 3.11 and
will be written there. Since our notation is already heavily burdened, we write only
U(M, P, p) until the end of the proof of Theorem 3.4.
Since each dominating all h
t,

, t T

, and dominating h
p,Hn
, n , gives
an (M, Q
T
)-generic G, the generic player can play fullling all theses largeness
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


SPECIALISING ARONSZAJN TREES AND PRESERVING SOME WEAK DIAMONDS 13
requirements and thereafter any

can be used as an argument of B. We use

this option to build a game between two players, and to establish properties that
say: The

-larger the argument in the Borel function B

1
is, the better it aims
at the envisaged weak diamond. See also the remark [15, V, Remark 5.4 (2)] about
the inuence of the guessed parameters on the generic lter. The knowledge that
the

-larger parameter can be inserted in the Borel function B

1
will help us later
to see that in the iteration every name of a real (called B

in Lemma 3.10 as it is
another Borel function) can be forced into a slalom from the ground model (called
( there) that is meagre and of Lebesgue measure zero.
The following theorem is an iterated version of Lemma 3.3. It is related to
Theorem 2.20, however now we want to compute bounded (M, P

)-generic lters
(that witness that no reals are added) as Borel functions of certain arguments.
As in Theorem 2.20 we use <
1
-properness and a tower M
i
: i ) with
= otp(M ) <
1
, = iteration length, of elementary submodels in order to
prove facts about M = M
0
and P

. The tower appears only in the proof, not in the

statement of the theorem. The following theorem would work for arbitrary iteration
length, but we use it only for length
2
and notate it only for this length.
Theorem 3.4. Let P
2
= P

, Q

:
2
, <
2
) be a countable support
iteration of iterands of the form Q
T
. If is suciently large and regular and if
M (H(), , <

) is countable and
(a) P

M,
2
,
(b) p P

M,
(c) = otp(M ),
(d) Let

be conal in M
1
. Let c: M be a bijection with c(0) = P

,
c(1) = p, c(2n + 2) =
n
, and let
U = U(M, P

, p)
= 2e(n
1
, n
2
) : c(n
1
) c(n
2
) 2e(n
1
, n
2
) + 1 : c(n
1
) <

c(n
2
).
Then there is a Borel function B = B

: (

T() T(), such that in the

following game
(M,P,p)
the generic player has a winning strategy , which depends
only on the isomorphism type of (M, , <

, P

, p,

):
() a play lasts moves,
() in the -th move the generic player chooses some real

and the antigeneric

player chooses some

, such that

,
() in the end the generic player wins i the following is true:
G

= c(n) : n B

: < ), U) is (M, P

)-generic and
p G

and
(q P

)(p q and q bounds G

).
Proof. We follow Abrahams exposition in [1, Theorem 5.17]. This theorem works
only inductively: For Q

in V
P
to be Dcomplete with respect to a system that
lies in V we need that P

does not add new countable sets of ordinals. So every

countable transitive set in V
P
is in V.
To prove the theorem we shall rst dene for every countable M (H(), ,
<

) with P

M, p P

M, with = otp(M ), an (M, P

)-generic lter
G

= B

(
i
: i < ), U); and then we shall prove that G

is bounded in P

by
a completely (M, P

)-generic condition. The bounding condition is not computed

in a Borel manner. Its existence is sucient, and its existence is proved along the
iteration.
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


14 HEIKE MILDENBERGER AND SAHARON SHELAH
Remark: The bounding condition also appears in an argument about the truth
in forcing extensions at the very end of our Lemma 3.11.
The denition of G

is by induction and we shall dene for every

0
< and
G
0
that is (M, P
0
)-generic and every p P

M with p
0
G
0
a lter G

that
extends G
0
and contains p. Once the induction is performed, we shall set
0
= 0,
G
0
= 0
P0
. There will be two main cases in this denition: successor and
limit, and likewise there will be two cases in the proofs that G

is bounded. We
start with the preparations for the successor case. When looking at complexity, we
regard G
0
as a parameter.
Two step iteration
Let P be a poset and let Q

V
P
be a name forced by 0
P
to be a poset.
Let be suciently large and regular (as said, = (2
2
)
+
is always suciently
large) and M
0
(H(), , <

) be a countable elementary submodel such that P,

Q

M
0
. Henceforth we write just H() instead of (H(), , <

). We want to nd
a criterion for when a condition (q
0
, q
1
) P Q

is completely (M
0
, P Q

)-generic.
Let : M
0
N
0
be a transitive collapsing map. Suppose that q
0
P is completely
generic over (M
0
, P) and let G
0
P M
0
be the (M
0
, P)-generic lter induced by
q
0
. Then (
0
=

G
0
is an (N
0
, (P))-generic lter and we can form the transitive
extension N

0
= N
0
[(
0
]. (Q

) is a name in N
0
, and its interpretation Q

0
= (Q

)[(
0
]
is a poset in N

0
.
Let G

V
P
be the canonical name of the P-generic lter over V. If F is a
(V, P) generic lter containing q
0
then M
0
[F] H()[F] can be formed and the
collapsing map on M
0
can be extended to collapse M
0
[F] onto N

0
. Let

be the
name of the extended collapse. Then q
0

P

: M
0
[G

] N

0
. We phrase now the
desired criterion and we shall use the direction from right to left later.
Lemma 3.5. Using the above notation, (q
0
, q
1
) is completely generic over (M
0
, P
Q

), i
1. q
0
is completely (M
0
, P)-generic, and
2. for some (
1
Q

0
that is (N

0
, Q

0
)-generic q
0

1
(
1
is bounded by q
1
.
In this case the lter induced by (q
0
, q
1
) over M
0
P Q

is
1
(
0
(
1
.
Given a countable M
0
H() such that the two step iteration P Q

is in M
0
,
our aim is to extent each (M
0
, P)-generic lter G
0
to an (M
0
, P Q

)-generic lter.
This denition depends not only on M
0
but also on another countable elementary
submodel M
1
H() such that M
0
M
1
and G
0
M
1
. In addition we x a
p
0
P Q

which we want to include in the extended lter. All of this leads us to

a ve place function E(M
0
, M
1
, P Q

, G
0
, p
0
) that we dene now.
Denition 3.6. Let P be a poset that adds no new countable sets of ordinals and
suppose that Q

, D

V
P
are such that

P
D

V is an
1
-completeness system and
Q

is D-complete with respect to D

.
Let be suciently large and M
0
M
1
(H(), , <

) be countable elementary
submodels with M
0
M
1
and P, Q

, D

M
0
. Let G
0
M
0
P be (M
0
, P)-generic
and suppose that G
0
M
1
. Let p
0
P Q

M
0
be given p
0
= (a, b

) with a G
0
.
Then we dene
G = E(M
0
, M
1
, P Q

, G
0
, p
0
),
an (M
0
, P Q

)-generic lter containing p

0
(dominating G
0
) by the following proce-
dure:
Let : M
1
N
1
with (M
0
) = N
0
be the transitive collapse and (
0
=

G
0
.
Form N

0
= N
0
[(
0
]. Observe that N

0
N
1
. Let Q

0
= (Q

)[(
0
], and let D
0
=
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


SPECIALISING ARONSZAJN TREES AND PRESERVING SOME WEAK DIAMONDS 15
(D

)[(
0
]. Then D
0
N
0
, because it is forced to be in the ground model. So D
0
=
(D) where D M
0
is a countably closed completeness system. Thus D
0
(N

0
, Q

0
, b

)
is dened in N
1
, where b

= (b

)[(
0
] is a condition in Q

0
. Since N
1
D
0
(N

0
, Q

0
, b

)
is countable,
(3.3) there is some (
1

(N
1
D
0
(N

0
, Q

0
, b

)).
(
1
is (N

0
, Q

0
)-generic and b

(
1
. Form (
0
(
1
= (, an (N
0
, (P Q

))-generic
lter. Then (p
0
) (. Finally we dene
(3.4) G = E(M
0
, M
1
, P Q

, G
0
, p
0
) =
1
(.
Now observe that if fulls Equations (3.1) and (3.2) for (N

0
, Q

0
, b

) instead of
(M, Q
T
, p), then the existence of Equation (3.3) is given by

1
(
1
= B
1
(, U(M
0
[G
0
], Q
0

[G
0
], b

[G
0
]))
and hence is Borel computable from and the code U of the intermediate model
(N

0
, Q

0
, b

).
In fact, we want to dene a formula so that
H() [= (M
0
, M
1
, P Q

, G
0
, p
0
)
i Equation (3.4) holds. That is, we want to dene E in H(). We cannot take
the above denition verbally, because it relies on the assumption that M
0
and M
1
are elementary substructures of H(), something which is not expressible in H().
Whenever the denition above relies on some fact that happens not to hold we let
( have an arbitrary value. For example if N

0
is not in N
1
or if N
1
D
0
(N

0
, Q

0
, b

)
is empty, then we let ( be some arbitrary xed N
0
-generic lter. The Borel compu-
tation does not invoke N
1
, since
1
(
1
= B
1
(, U(M
0
[G
0
], Q
0

[G
0
], b

[G
0
])). Here,
G
0
is a parameter and will be set 0
P0
later, so that in the end (that means in
Lemma 3.11) only the possible isomorphism types of (M
0
, M
0
, <

M
0
, P

, p,

)
need to be guessed stationarily often alongside with names for the F and f from
the statement of the weak diamond.
The following lemma shows the second part of the argument: We want to show
the G given in Equation (3.4) is bounded. The lemma analyses the iteration of two
posets when the second is D-complete.
Lemma 3.7. The One Step Extension Lemma. Let P be poset and suppose that
Q

, D

V
P
are such that

P
D

V is an
1
-completeness system and
Q

is D-complete with respect to D

.
Let be suciently large and M
0
M
1
H

be countable elementary submodels

with M
0
M
1
and P, Q

, D

M
0
. Suppose that q
0
P is (M
1
, P)-generic as
well as completely (M
0
, P)-generic, and let G
0
M
0
P be the M
0
lter over
M
0
P induced by q
0
. Let p
0
P Q

, p
0
M
0
be given, so that p
0
= (a, b

)
and a G
0
. Then there is q
1
V
P
such that (q
0
, q
1
) is completely generic over
(M
0
, P Q

) and p
0
(q
0
, q
1
), in fact (q
0
, q
1
) bounds G = E(M
0
, M
1
, P Q

, G
0
, p
0
) =
G
0
B
1
(, U(N

0
, Q

0
, (b

))).
Proof. This is literally [1, The Gambit Lemma]. For completeness sake we repeat
Abrahams proof here. Notice that G
0
M
1
by the following argument: Let R be
the collection of all conditions r P that are completely generic over M
0
. Then
R M
1
and q
0
RM
1
. Since q
0
is (M
1
, P)-generic, it follows that it is compatible
with some r R M
1
. But any two compatible conditions in R induce the same
lter, and hence G
0
is the lter induced by r.
Let : M
1
N
1
, (M
0
) = N
0
, be the transitive collapse and (
0
=

G
0
.
We recall the denition of E(M
0
, M
1
, P Q

, G
0
, p
0
). Form N

0
= N
0
[(
0
] and let
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


16 HEIKE MILDENBERGER AND SAHARON SHELAH
Q

0
= (Q

)[(
0
], and let D
0
= (D

)[(
0
]. Then D
0
N
0
because it is forced to
be in the ground model. So D
0
= (D) where D M
0
is a countably closed
completeness system. Thus D
0
(N

0
, Q

0
, b

) is dened in N
1
, where b

= (b

)[(
0
]
is a condition in Q

0
. Since N
1
D
0
(N

0
, Q

0
, b

) is countable, there is some (

1

(N
1
D
0
(N

0
, Q

0
, b

)). (
1
is (N

0
, Q

0
)-generic and b

(
1
. Form (
0
(
1
= (,
an (N
0
, (P Q

))-generic lter. Then (p

0
) (. We dened G = E(M
0
, M
1
, P
Q

, G
0
, p
0
) as
1
(.
Let G

V
P
be the canonical name of the generic lter over P. Then q
0
forces
that can be extended to a collapse

which is onto N

0
, that is
q
0

P

: M
0
[G

] N

0
.
The conclusion of our lemma follows if we show that
(3.5) q
0

P

1
(
1
is bounded in Q

.
In this case, if we dene q
1
V
P
so that q
0

P
q
1
bounds

1
(
1
, then the
previous lemma implies that the (M
0
, P Q

)-generic lter induced by (q

0
, q
1
) is

1
(
0
(
1
.
So let F be (V, P)-generic with q
0
F.

[F] collapses M
0
[F] onto N

0
and there
is a set X D
0
(N

0
, Q

0
, b

), so that if H X is any lter then

1
H is bounded
in Q

[F]. As N
1
[F] H

[F], we can have X N

1
[F]. But since D
0
is in the ground
model, X N
1
. Thus (
1
X, where (
1
is the lter dened above. This proves
Equation (3.5).
The iteration theorem
Let P

be a countable support iteration of length obtained by choosing iterands

Q

V
P
as in the theorem. That is, each Q

is D-complete in V
P
for some
1
-
completeness system taken from V. Let be a suciently large regular cardinal.
To prove the theorem we rst describe a machinery for obtaining generic lters over
countable submodels of H(). We dene a function E that takes ve arguments,
E(M
0
,

M [1, ), P

, G
0
, p
0
) of the following types.
1. M
0
H

is countable, P

M
0
, so M
0
. Moreover, p
0
M
0
P

.
2. For some
0
M
0
, G
0
is an (M
0
, P
0
)-generic lter and such that
p
0

0
G
0
. We assume that G
0
M
1
.
3. The order type of M
0
[
0
, ) is .
4.

M = M

: 0 ) is an +1-tower of countable elementary submodels

of H() and M
0
= M. Note that only M
0
= M appears in the statement
of the theorem. The rest M

: 1 ) of the tower is a technical

means for the proof.
The value returned, G

= E(M
0
,

M [1, ), P

, G
0
, p
0
) is an (M
0
, P

)-generic
lter that extends G
0
and contains p
0
. Formally, in saying that G

extends G
0
, we
mean that the restriction projection takes G

onto G
0
. The denition of E(M
0
,

M
[1, ), P

, G
0
, p
0
) is by induction on <
1
.
Assume that =

+1 is a successor ordinal. Then =

+1 is also a successor.
Assume rst that
0
=

. Then = 1 and we have only two structures: M

0
and
M
1
. Since P

is isomorphic to P
0
Q
0
we can dene G

by Equation (3.4). So,

if fulls Equations (3.1) and (3.2) for (M
0
[G
0
], Q
0

[G
0
], b

[G
0
]) in the role of of
(M, Q
T
, p), then
G

= E(M
0
, M
1
, P
0
Q
0
, G
0
, p
0
) = G
0
B
1
(
0
, U(M
0
[G
0
], Q
0

[G
0
], b

[G
0
])).
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


SPECIALISING ARONSZAJN TREES AND PRESERVING SOME WEAK DIAMONDS 17
Assume next that
0
<

. Then by induction hypothesis, if all

i
, i <

, are
suciently large, then
G

=E(M
0
, M

: 1

), P

, G
0
, p
0

)
=G
0
B

(
i
: 0 i <

), U(M
0
[G
0
], P
[0,

[G
0
], p
0
[
0
,

[G
0
]))
(3.6)
is dened and is an (M
0
, P

)-generic lter that extends G

0
and contains p
0

.
Moreover by elementarity, G

. When we nish this denition it will be

evident that it continues for every <
1
since M

H() and the parameters

are all in M

. This brings us to the previous case and we choose

such that it
fulls Equations (3.1) and (3.2) in which (M, Q
T
, p) is replaced by
(M
0
[G

], Q

[G

], p
0
(

[G

]).
Now from Equation (3.6) we dene temporarily
(3.7) U

= U(M
0
[G
0
], P
[0,

[G
0
], p
0
[
0
,

[G
0
)]).
Then
G

=E(M
0
, M

, P

, G

, p
0
)
=G
0
B
1
(

, U(M
0
[G
0
B

(
i
: i <

), U

)],
Q

[G
0
B

(
i
: i <

), U

)],
p
0
(

[G
0
B

(
i
: i <

), U

)]))
=: G
0
B

(
i
: i < ), U(M
0
[G
0
], P
[0,)

[G
0
], p
0

[G
0
]))
(3.8)
and the middle U

is dened above in Equation (3.7). This justies that the Borel

functions given by induction hypothesis can be composed to one Borel function of
the required arguments.
Now it is also clear how to dene the strategy (
i
,
i
: i <

)): The generic

player plays

so that it fulls Equations (3.1) and (3.2), where (M, Q

T
, p) is
replaced by (M
0
[G

], Q

[G

], p
0
(

[G

]) with G

as in Equation (3.6).
Now assume that is a limit ordinal and let
n
: n ) be an increasing
conal sequence with
0
= 0. Let
n
M
0
be such that
n
= otp(M
0
[
0
,
n
)).
Let I
n
: n ) be an enumeration of all dense subsets of P

that are in M
0
in such
a way that I
n
is the <

-least dense subset of P

that is not among I

m
: m < n.
We dene
G

=E(M
0
,

M [1, ), P

, G
0
, p
0
)
=G
0
B

(
i
: i < ), U(M
0
[G
0
], P
[0,)

[G
0
], p
0
[
0
, )

[G
0
]))
as follows. We dene by induction on n a condition p
n
P

M
0
and an
(M
0
, P
n
)-generic lter G
n
M
n+1
such that
1. G
0
and p
0
are given. p
n

n
G
n
.
2. p
n
p
n+1
and p
n+1
I
n
.
Suppose that G
n
and p
n
are dened. First we can nd p
n+1
I
n
M
0
such that
p
n+1

n
G
n
(for an existence proof see [1, Lemma 1.2]) and we take the <

-least
in M
0
so that it is Borel computed. Now dene
G
n+1
=E(M
0
, M

:
n
+ 1
n+1
), P
n+1
, G
n
, p
n+1

n+1
)
=G
0
B
n+1n
(
i
: i [
n
,
n+1
)), U

)
Here we have
U

=U(M
0
[G
0
B
n
(
i
: i <
n
), U

)],
P
[n,n+1)

[G
0
B
n
(
i
: i <
n
), U

)],
p
n+1
[
n
,
n+1
)

[G
0
B
n
(
i
: i <
n
), U

)]) and
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


18 HEIKE MILDENBERGER AND SAHARON SHELAH
U

= U(M
0
[G
0
], P
[0,n)

[G
0
], p
n+1
[
0
,
n
)

[G
0
]).
Finally let
G

= the generic lter generated in M

0
by p
n
: n .
From the above induction on n < and from the induction hypothesis it is clear
that there is a Borel function B

such that
(3.9) G

= G
0
B

(
i
: i < ), U(M
0
[G
0
], P
[0,)

[G
0
], p
0
[
0
, )

[G
0
])).
This ends the denition of E(M
0
,

M [1, ), P

, G
0
, p
0
) and of B

.
The strategy for the generic player is dened by the prescription, that in the
limit game of length he plays according to the strategies for the initial segments
of the game. (This justies that

is just named , for all lengths .) This is a

winning strategy, as the Borel function was just derived. It gives a generic lter.
We still have to show that the given generic lter is bounded.
Now the missing part is to show that all the generic lters are bounded is
preserved in the limit steps of the iteration. Again there is nothing new to our work
and we repeat Abrahams proof to [1, The Extension Lemma].
Lemma 3.8. Let P

, Q

: < , ) be a countable support iteration of

forcing posets such that each iterand Q

satises the following in V

P
:
1. Q

is -proper for every countable .

2. Q

is D-complete with respect to some countably closed completeness sys-

tem in the ground model that has the property that all

serve as
parameters.
Suppose that M
0
H() is countable, P

M
0
and p
0
P

M
0
. For any

0
M
0
with = otp(M
0
[
0
, )) and

M = M

: ) is a tower of
countable elementary substructures starting with the given M
0
, then the following
holds:
For every q
0
P
0
that is completely (M
0
, P
0
)-generic as well as (

M, P
0
)-
generic, if p
0

0
< q
0
, then there is some q P

such that q
0
= q
0
and
p
0
< q and q is completely (M
0
, P

)-generic. In fact, the lter induced by q is

E(M
0
, M

: 1 ), P

, G
0
, p
0
) where G
0
P
0
M
0
is the lter induced by
q
0
.
Proof. Let G
0
P
0
M
0
be the M
0
-generic lter induced by q
0
. Observe that
G
0
M
1
follows from the assumption that q
0
is also M
1
-generic. We shall prove
by induction on = otp(M
0
[
0
, )) that q can be found that bounds G

=
E(M
0
, M

: 1 ), P

, G
0
, p
0
).
Suppose rst that =

+1 and consequently =

+1 are successor ordinals.

Dene in M

, X P
0
as maximal antichain of conditions r so that
1. r bounds G
0
,
2. r in M

: 1

)-generic.
Then X M

is predense above q
0
. By our inductive assumption, every r
0
X
has a prolongation r
1
P

that bounds G

= E(M
0
, M

: 1

), G
0
, p
0

). Since all the parameters are in M

, we get that G

. Since M

H()
we can choose r
1
M

whenever r
0
X M

. This denes a name r

1
V
P
0
,
forced by q
0
to be in M

. Namely, if G is any (V, P

0
)-generic lter containing
q
0
, then X G contain a unique condition r
0
, and we let r

1
[G] = r
1
. By the
Properness Extension Lemma [1, Lemma 2.8] we can nd q
1
P

, q
1

0
= q
0
, q
1
is (M

, P

)-generic, and q
1

P

1
is in the generic lter G

. It follows that q
1
bounds G

. We nd q
2
P

, such that q
2

= q
1
and q
2
bounds G

. In order
to dene q
2
() we use the Two Step Lemma and Equation (3.5).
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


SPECIALISING ARONSZAJN TREES AND PRESERVING SOME WEAK DIAMONDS 19
Now assume that is a limit ordinal. We follow the denition of G

see Equa-
tion (3.9). Recall that we had an -sequence
n
: n ) conal in and we
dened
n
conal in as the resulting sequence
n
= otp(M
0
[
0
,
n
)). We de-
ned by induction p
n
P

M
0
and lters G
n
P
n
, G
n
M
n+1
and dened
G

as the lter generated by the p

n
s. We shall dene now q
n
P
n
by induction
on n so that the following hold
1. q
n
bounds G
n
,
2. p
n

n
q
n
,
3. q
n
= q
n+1

n
,
4. q
n
is M

:
n
+ 1 )-generic over P
n
.
Thus q
n
gains in length and looses in status as an M

-generic condition for 0 <

n
. But q
n
is completely (M
0
, P
n
)-generic for all n. Finally q =

q
n
is not
M

-generic for any > 0. However, q is completely (M

0
, P

)-generic.
Suppose that q
n
is dened. Let X be in M
n+1+1
be a maximal antichain in P
n
of conditions r that induce G
n
and are M

:
n
+1
n+1
)-generic over P
n
.
Observer that X is predense above q
n
. For each r
0
X, dene by the induction as-
sumption r
1
P
n+1
such that r
1
bounds G
n+1
, p
n+1

n+1
< r
1
and r
1

n
= r
0
.
If r
0
X M
n+1+1
, then r
1
is taken from M
n+1+1
. Now view r
1
: r
0
X
as a name r

for a condition forced by q

n
to lie in M
n+1+1
. By the -Extension
Lemma [1, Lemma 5.6], dene q
n+1
that satises items 2 to 4 from the above list
and such that q
n+1

P
n+1
r

G
n+1

. Then q
n+1
bounds G
n+1
and is a required.
End of proof of Theorem 3.4: Now that the induction is performed, we set

0
= 0, G
0
= 0
P0
, p
0
= p P

from the statement of Theorem 3.4. Then

N

0
= N
0
= (M
0
), (P
[0,)

)[(
0
] = (P

) and (p
0
)
[0,)

)[(
0
] = (p) and the B

s
second argument is just the isomorphism type of (M
0
, , <

, P

, p,

).
3.4
The role of the antigeneric player in the game (M, P

, p) is now turned to good

use:
Denition 3.9. Let f, g V

. A notion of forcing P

has the (f, g)-bounding

property when for every P

-name u

for a function from to the following holds:

If p
P
u

g, then there are q p and an f-slalom S

n
: n < ) in the ground
model such that q
P
(n)(u

(n) S
n
). S
n
: n < ) is an f-slalom if for every
n, S
n
and [S
n
[ f(n).
Lemma 3.10. Suppose that
() <
1
, and
() B

is a Borel function from (

to 2

,
() r : , r diverging to innity, and lim
r(n)
2
n
= 0.
Then we can nd some ( = (
B
such that
(a) ( is a closed subset of 2

,
(b) (n)[ n : ([ r(n), so if ( = lim(T) = f 2

: nf n T,
then T 2
<
is a tree with n-th level counting less than or equal to r(n),
(c) in the following game
(,B

)
between two players, IN and OUT, the player
IN has a winning strategy, the play lasts moves and in the -th move
OUT chooses

and then IN chooses

. In the end IN wins

i B

: < )) (.
Proof. Assume that P

= P

, Q

: , < ) is a c.s. iteration of Laver forcing

and assume that p P

and

: < ) is a sequence of names for the P

-generics.
Clearly p
P

: < )) 2

.
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


20 HEIKE MILDENBERGER AND SAHARON SHELAH
The (f, g)-bounding property is preserved in countable support iterations [3, p.
340]. The Laver forcing and any forcing not adding reals at all have the (f, g)-
bounding property. Hence there are p

and ( as in (a) and (b) above such

that
p

: < )) (.
Now we show that player IN can play in a way that imitates the Laver-generic reals
over a countable elementary submodel, so that actually everything is in the ground
model.
Let M

(H(), ) be countable such B

, ( M

. (So M

is not the M from

the next proof, but rather contains a non-trivial part of the power-set of that M.)
Now we prove by induction on j for all i < j

i,j
Assume that P

j
M

and G
i
P

i
M

is generic over M

, and p

is such
that p

j
M

and p

i G
i
. Then in the following game

(i,j,Gi,p

)
player II has a winning strategy
(i,j,Gi,p

)
. There are j i moves indexed
by [i, j), and in the -th move (p

) are chosen such that player

I chooses p

/G
i
, p

, and

and player II chooses

.
First case: there is a (P

, M

)-generic G

, such that p

()
G

and G

G
i
and ( [i, )

[G

] =

and M

[G

] [= p

().
In this case player I chooses p

forcing this and so that M

[G

] [=
p

()
P

. Then player I chooses

dominating M

[G

] and the second

player chooses

.
Second case: There is no such G

. Then player I won the play.

We prove by induction on j that player II wins the game

(i,j,Gi,p

)
: Case 1:
j = 0. Nothing to do. Case 2: j = j

+ 1. For [i, j) we use the strategy for

(i,j,Gi,p

)
, and for = j we make the following move: We show that there is a
generic G
j

of Q

[G
j
]
j
to which p

(j

) belongs and such that

j
[G
j

j
.
Then the move

j
[G
j

] dominates

[G
j
] and also player Is move
j
.
First take q p

(j

) such that q is (M

[G
j
], Q

[G
j
]
j
)-generic. q V is a
Laver condition. Now we take a stronger condition q

by letting trunk(q) = trunk(q

)
and for every s q

of length n,
succ(q

, s) = n succ(q, s) : n
j
(n).
Now let G
j

= r M

[G
j
] : q

r. Since q

is a (M

[G
j
], Q

[G
j
]
j
)-
generic condition, G
j

is a (M

[G
j
], Q

[G
j
]
j
)-generic lter. The generic real is

j
[G
j

] =

trunk(p) : p G
j

. Then q

j
. Now player II takes

j
=

j
[G
j

]. We set G
j
= G
j
G
j

. Case 3: j is a limit. Like the proof of the

preservation of properness.
Why does
i,j
suce? Use i = 0, j = , B

. Take P

, p

.
Let (0, , , p

) be a winning strategy for player II in the game

(0,,{},p

)
.
During the play of
(,B

)
let

be chosen in stage < . The player IN simulates

on the side a play of

(0,,{},p

)
: As a move of I he assumes the

chosen by OUT
in the play of
(,B

)
and p

, p

= p

for < , the p

gotten from earlier

simulations. Then player IN uses (0, , , p

) for player II, applied to (p

), to
compute an

, which he presents in this move in

(,B

)
. So p

forces that there

is a Laver generic

[G

] =:

over M

[G

] and that

. The requirement

is fullled.
Suppose that they have played. So we have

: < ) and there is

p =

<
p

, and for < there is the name for the Q

-generic real, namely

8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


SPECIALISING ARONSZAJN TREES AND PRESERVING SOME WEAK DIAMONDS 21

, such that for all < , p

P

. So as p
P

: < ))
(, we have B

: < )) (.
Let S
1
be stationary and A

: S) exemplify
S
. For example we can
take the most frequent S = <
1
: limit ordinal, which gives
1
.
Lemma 3.11. Let r : such that lim
r(n)
2
n
= 0. Assume that V [=
S
. Then

P
2
(2

, lim(T) : T R perfect (n)[ n : lim(T)[ r(n), ).

Proof. Let G be P
2
-generic over V. We use the
S
-sequence A

: S) in the
following manner: By easy integration and coding we have (N

, f

, F

, C

, P

2
,
p

, <

) : S) such that
(a) N

is a transitive collapse of a countable M H(, , <

), <

is a well-
ordering of N

, U

codes the isomorphism type of (N

, P

2
, p

).
(b) N

[= P

2
= P

, Q

:
N

2
, <
N

2
) is as in Denition 2.10.
(c) N

[= (p

2
, f

is a P

2
-name of a member of
1
2 F

: 2
<1
2

).
(d) If p P
2
,
p
P
2
f

2
1
F

: 2
<1
2

is Borel, C

1
is club,
and p, P
2
, F

, f

, C

H(), then
S(p, F

, f

) := S : there is a countable M (H(), , <

)
such that f

, F

, C

, P
2
, p M
and there is an isomorphism h

from N

onto M
mapping P

2
to P
2
, f

to f

,
F

to F

, C

to C

, p

to p, <

to <

M
is a stationary subset of
1
.
(e) Choose B
()
: S) such that () = otp(N

2
) and
B
()
: (

)
()
T() Gen
+
(P

2
)
= G P

2
N

: G is P

2
-generic over N

and bounded
be as in Theorem 3.4 with U

= U(N

, P

2
, p

).
We do not require uniformity,

: < ()) is indeed

: < ())
since we have the dependence on the in the denition of B
()
. We assume that
N

1
= . Since this holds on a club set of
1
, this is no restriction.
Now assume the p G and F

, f

, C

are as in (d).
We dene a function B

,U

with domain (

)
()
.
B

,U
(

: < ())) =
_
_
_
F

(f

)[B
()
(

: < ()), U

)], if the argu-

ment is good;
0, 0, . . . , ) 2

, otherwise.
Here, we call

: < ()) a good argument if there is a play

: <
()) in the game
(N

,P

,p

)
from Theorem 3.4 in which the generic player plays
according his winning strategy and the antigeneric player plays according to the
rules. Goodness is a Borel predicate because the

are irrelevant, just check whether

the

are large enough for Equations (3.1) and (3.2) in the respective iteration
step. So B

,U

: < ()) is a Borel function. Now we choose a very good

argument

: < ()) that player IN plays with his strategy in

((),B

,U

)
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


22 HEIKE MILDENBERGER AND SAHARON SHELAH
from Lemma 3.10 applied to B

,U

and the (r, 2

n
) bounding property, answering to
a good argument

: < ()) played by player OUT.

Now we derive a guessing function g. We consider for every S a very good
argument

: < ()). We assume that G is P

2
-generic over V and that p G.
Then we also have by the rules of the game
(N

,P

,p

)
that
B
()
(

: < ()), U

) has an upper bound q

.
Lemma 3.10 gives a closed set (
B

,U

with small levels as in 3.10(b), such that for

S, and we have
(3.10) B

,U

: < ())) (
B

,U

.
Note that (
B

,U

does not depend on

: < ()). So (3.10) also holds for

: < ()) that are the answers of player IN in the game

((),B

,U

)
to any
good sequence

: < ()) given by the generic player that is so fast growing

that B
,U

: < ())) computes a bounded generic lter over M as in

Theorem 3.4. This is important, since the isomorphism h

does not preserve the

knowledge (that is which branches are continued and what are the values of the
promises in these continuations) about the level for the Aronszajn trees involved
in P M.
We set
(
B

,U

=: g().
Both sides are conceived as Borel codes for closed sets. Since M and N

we have that h

((
B

,U

) = (
B

,U

. We show that g is a diamond function.

Since P
2
is proper, S(p, f

, F

) is also stationary in V[G]. Now we take a very

good sequence

: < ()) that is suitable so that B

,U

: < ()))
computes a bounded (M, P)-generic lter for M that witnesses that S. So now
we take the game
(M,P,p)
for the choice of the

: <

) and then again we take

the winning strategy in the game
((),B

,U

)
, which is unchanged by the collapse,
for choosing

: <

). We take q to be a bound of B
,U

: < ())).
Now we have that q p and
q B
()
(

: < ()), U

) is (M, P)-generic and bounded by q.

Now for S(p, f

, F

) we have by the isomorphism property of h

and by (3.10),
q h

(f

) = F

(f

) F

(f

) g() C

.
So we have that p forces that S : F(f ) g() contains a stationary subset
of S(p, f

, F

). Note that the stationary subset depends on F (and f of course), but

the guessing function g does not. So actually we proved a diamond of the kind:
There is some g :
1
B such that for every Borel map F : 2
<1
A and for
every f :
1
2 the set

1
: F(f )Eg()
is stationary.
Corollary 3.12.
P
2
(R, ^, ).
Proof. Leb(g()) = 0 for the functions g :
1
closed subsets of 2

from the
previous lemma. Thus, for every Borel F : 2
<1
2

, the function g :
1
^
is a guessing sequence showing
P
2
(R, ^, ), and we nish the proof of Theo-
rem 1.2.
1.2
8
4
8


r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
9
-
0
8







m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
9
-
1
0


SPECIALISING ARONSZAJN TREES AND PRESERVING SOME WEAK DIAMONDS 23
Since ( from Lemma 3.10 is also meagre, the same proof also yields
Corollary 3.13.
P
2
(R, /, ).
If S
1
is stationary and we start with
S
in the ground model, then we
get the respective weak diamonds on S. We conclude with an open question: The
forcing from Denition 2.10 could easily be mixed with proper
2
-p.i.c. iterands, for
example iterands with [Q

[
1
(by [15, Lemma VIII 2.5] this is sucient for the

2
-p.i.c.) that add reals. Still we specialise all Aronszajn trees in the new mixed
iteration. However, the parallel of our main technique for the weak diamonds, that
is Theorems 3.4 and 3.11, does not work anymore, since the completeness systems
are no longer in the ground model. So there is the question:
Question 3.14. Is 2
0
=
2
and (Cov(^)) and all Aronszajn trees are special
consistent relative to ZFC?
References
[1] Uri Abraham. Proper forcing. In Matthew Foreman, Akihiro Kanamori, and Menachem Magi-
dor, editors, Handbook of Set Theory. Kluwer, To appear.
[2] Uri Abraham and Saharon Shelah. A
2
2
well-order of the reals and incompactness of
L(Q
MM
). Annals of Pure and Applied Logic, 59:132, 1993.
[3] Tomek Bartoszy nski and Haim Judah. Set Theory, On the Structure of the Real Line. A K
Peters, Wellesley, Massachusetts, 1995.
[4] James Baumgartner, Leo Harrington, and Eugen Kleinberg. Adding a closed unbounded set.
J. Symbolic Logic, 41:481482, 1976.
[5] Andreas Blass and Saharon Shelah. There may be simple P

1
- and P

2
-points and the Rudin-
Keisler ordering may be downward directed. Ann. Pure Appl. Logic, 33:213243, [BsSh:242],
1987.
[6] Keith Devlin and Howard Johnsbr
o
aten. The Souslin Problem, volume Lecture Notes in Math-
ematics. Springer, 1974.
[7] Keith J. Devlin and Saharon Shelah. A weak version of which follows from 2

0
< 2

1
.
Israel J. Math., 29:239247, 1978.
[8] James Hirschorn. Random trees under CH. Israel J. Math., 157:123154, 2007.
[9] Ronald B. Jensen. The ne structure of the constructible hiercharchy. Annals Math. Logic,
4:229 308, 1972.
[10] Alexander Kechris. Classical Descriptive Set Theory. Springer, 1995.
[11] Richard Laver. Random reals and Souslin trees. Proc. Amer. Math. Soc., 100(3):531534,
1987.
[12] Heike Mildenberger. Creatures on
1
and weak diamonds. To appear in The Journal of Sym-
bolic Logic.
[13] Arnold Miller. Arnie Millers problem list. In Haim Judah, editor, Set theory of the reals
(Ramat Gan, 1991), volume 6 of Israel Math. Conf. Proc., pages 645654. Bar-Ilan Univ.,
Ramat Gan, 1993.
[14] Justin Tatch Moore, Michael Hrus ak, and Mirna Dzamonja. Parametrized -principles. Trans.
Amer. Math. Soc., 356:22812306, 2004.
[15] Saharon Shelah. Proper and Improper Forcing, 2nd Edition. Springer, 1998.
[16] Peter Vojt as. Generalized Galois-Tukey connections between explicit relations on classical
objects of real analysis. In Haim Judah, editor, Set Theory of the Reals, volume 6 of Israel
Mathematical Conference Proceedings, pages 619643, American Mathematical Society, 1993.
Heike Mildenberger, Universit at Wien, Kurt G odel Research Center for Mathemat-
ical Logic, W ahringer Str. 25, 1090 Vienna, Austria
Saharon Shelah, Institute of Mathematics, The Hebrew University of Jerusalem,
Givat Ram, 91904 Jerusalem, Israel, and Mathematics Department, Rutgers University,
New Brunswick, NJ, USA
E-mail address: heike@logic.univie.ac.at
E-mail address: shelah@math.huji.ac.il