Professional Documents
Culture Documents
9
0
r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
0
m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
1
Complete Quotient Boolean Algebras
Akihiro Kanamori and Saharon Shelah*
For I a proper, countably complete ideal on T(X) for some set X, can the quotient Boolean algebra
T(X)/I be complete? This question was raised by Sikorski [Si] in 1949. By a simple projection argument as
for measurable cardinals, it can be assumed that X is an uncountable cardinal , and that I is a -complete
ideal on T() containing all singletons. In this paper we provide consequences from and consistency results
about completeness. Throughout, will denote an uncountable cardinal, and by an ideal over we shall
mean a proper, -complete ideal on T() containing all singletons.
If is a measurable cardinal and I a prime ideal over , then of course T()/I is complete, being the
two-element Boolean algebra. The following theorem shows that completeness in itself has strong consistency
strength:
Theorem A. If
3
and there is an ideal I over such that T()/I is complete, then there is an
inner model with a measurable cardinal.
The restriction
3
is necessary for our proof.
There is a well-known situation in which completeness obtains. An ideal over is -saturated i
whenever X
, I. Smith-Tarski [ST]
established that if an ideal I over is
+
-saturated, then T()/I is complete. There are several consistency
results about the existence of such ideals; modifying a sharp one due to the second author, we establish
the following result separating completeness from saturation. NS denotes the ideal over
1
consisting of the
non-stationary subsets; for an ideal I over and A , I[A denotes X [ X A I, the restriction
of I to A; nally, for S
1
,
S denotes
1
S.
Theorem B. Suppose that is a Woodin cardinal, CH holds, and S is a stationary, co-stationary
subset of
1
. Then there is a cardinal-preserving forcing extension with no new reals satisfying: 2
1
= ,
and I = NS[
S is an ideal over
1
such that T(
1
)/I is a complete Boolean algebra (and I is not
2
-saturated).
Of course, starting with a Woodin cardinal we can insure that CH also holds by carrying out a pre-
liminary extension. Also, the statement has the same force if S is replaced by
S, but the formulation is
notationally convenient for the proof. That the I in the resulting model cannot be
2
-saturated follows from
the following well-known result of Ketonen [Ke]: If CH holds and there is an
2
-saturated ideal over
1
,
then 2
1
=
2
. Adding a further layer of complexity to the proof of Theorem B, we also establish:
* The second author is partially supported by the U.S.-Israel Binational Science Foundation. This paper
is #390 in his bibliography.
1
3
9
0
r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
0
m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
1
Theorem C. Assume the hypotheses of Theorem B. Then there is a forcing extension with no new reals
satisfying: 2
1
=
3
= , and I = NS[
S is an ideal over
1
such that T(
1
)/I is a complete Boolean algebra
(and I is not
2
-saturated).
It follows from 1.1(a) below that I must be
3
-saturated. This then contributes to the theory of saturated
ideals by establishing the relative consistency of 2
=
1
, 2
1
=
3
, and the existence of an
3
-saturated
ideal over
1
.
In 1 we derive some consequences of completeness and establish Theorem A. In 2 we indicate the
modications necessary to a previous proof of the second author to establish Theorem B. Finally in 3 we
build on 2 to establish Theorem C, providing iteration lemmas for iterated semiproper forcing with mixed
supports. The main mathematical advances in this paper are due to the second author, based on speculations
and prodding by the rst.
1 Consequences of Completeness
We rst review the various concepts involved to arm some notation: Let I be an ideal over . Then
I
+
= T() I, the positive measure sets with respect to I. For any A , [A]
I
= B [ A B
I. The Boolean algebra T()/I consists of the [A]
I
s with the set-theoretic operations modulated by
I : [A]
I
[B]
I
= [AB]
I
, [A]
I
[B]
I
= [AB]
I
, [A]
I
= [A]
I
, 0
I
= []
I
, and 1
I
= []
I
. The subscript I
will be suppressed when clear from the context. A Boolean algebra is complete i least upper bounds exist
for any collection of its members. / is an antichain with respect to I i / I
+
yet whenever A, B / are
distinct, A B I. Thus, I is -saturated i every antichain with respect to I has cardinality < . Also,
/ is a maximal antichain i
[A] [ A / = 1 yet whenever A, B / are distinct, [A] [B] = 0.
The rst signicant results bearing on Sikorskis question were derived by Solovay [So], who established
the consistency, relative to the existence of a measurable cardinal, of the existence of saturated ideals over
accessible cardinals. In passing, he in eect noted the following partial converse to the Smith-Tarski result:
1.1 Lemma. Suppose that I is an ideal over such that T()/I is complete. For any , if I is not
-saturated, then 2
. In particular,
(a) I is 2
-saturated.
(b) If 2
< 2
+
, then I is
+
-saturated.
Proof. Let A
[A
[ T()/I[ 2
.
Kunen established that if there is a
+
-saturated ideal over , then is measurable in an inner model.
2
3
9
0
r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
0
m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
1
In particular, as Solovay noted, 1.1(b) implies that if there is an ideal I over such that T()/I is complete,
then V ,= L. Kunen asked in the early 1970s whether completeness has strong consistency strength, and
Theorem A conrms this, at least if
3
.
With our ultimate goal the proof of Theorem A, we now x an ideal I over such that T()/I is
complete for the rest of this section. We use the well-known strategem of considering T()/I 0 as a
notion of forcing with [X] [Y ] i X Y I, and we denote the corresponding forcing relation simply by
|-. Note that if A
] |- = i = .
The following lemmata derive consequences of completeness using maximal antichains.
1.2 Lemma. Suppose that and is a name such that |- . Then there is a partition
B
I
+
, then [B
] |- = .
Proof. Let / be a maximal antichain with respect to I such that for any A /, there is a
A
<
such that [A] |- =
A
. By completeness, let A
] =
[A] [
A
= . Then
set B
= A
<
A
] = [A
] by -completeness. The B
).
For each < , let B
= sup < [ B
and < ,
so that
= < [
<
E
= is a
partition. Consequently, for any X I
+
there is an < such that E
X I
+
by -completeness and
[E
X] |-
.
The proof of the following proposition is similar; it will not be needed in the rest of the paper. The only
early result about complete quotient Boolean algebras other than the Smith-Tarski result, it appeared in
terms of distributivity properties in Pierce [P], which also contained a similar formulation of the easy forcing
fact that a notion of forcing adjoins a new function: exactly when it adjoins a new subset of .
1.4 Proposition. Suppose that 2
< . Then forcing with |- does not adjoin any new functions:
2
.
Proof. Suppose that |- : 2
[ < 2
and
its = (). For each f : 2
, let E
f
=
<
B
f()
. Then
f
E
f
= is a partition. Consequently, for
any X I
+
, there is an f such that E
f
X I
+
by -completeness, and [E
f
X] |- =
f.
3
3
9
0
r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
0
m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
1
The connection with inner models of measurability is made through the well-known concept of precipi-
tous ideal, due to Jech and Prikry. For an ideal J over , if G is generic over V for the corresponding notion
of forcing T()/J 0
J
, then X [ [X] G is an ultralter on T() V , and for any class A in the
sense of V , the ultrapower of A with respect to this ultralter using functions in
V V is called the generic
ultrapower of A by G. In this situation, if is a name for a function in
A V , we denote by ( ) the name
of the equivalence class of in the ultrapower. J is precipitous i for any such G, the generic ultrapower
of V by G is well-founded. As Jech and Prikry showed, if there is a precipitous ideal J over , then is
measurable in an inner model.
Before we turn to the proof of Theorem A, we establish a partial well-foundedness result about generic
ultrapowers. Continuing to work with our xed ideal I over such that T()/I 0 is complete and the
corresponding notion for forcing |-, we rst make an observation related to 1.2.
1.5 Lemma. Suppose that < and [X] |- :
V . Then there is an f: such that
[X] |- ( ) = (
f), i.e. [X] forces that and
f belong to the same equivalence class in the generic ultrapower.
Proof. For any set x, let c
x
denote the constant function: x. We rst note that for any
[Y ] [X], there is a [Z] [Y ] and a < such that [Z] |- ( ) = ( c
). (There is a [Y ] [Y ] and a g:
such that [Y ] |- = g. By -completeness there is a < such that Z = i Y [ g(i) = I
+
. Thus,
[Z] |- ( ) = ( c
] =
[A] [
A
= .
Then as in the proof of 1.2, let B
] = [A
] and B
[ < is a partition
of X. Finally, dene f: by f(i) = i i B
[ <
1
) is a sequence of maximal antichains with respect to I, for any X I
+
there is a Y T(X) I
+
such that for any <
1
, [A /
[ A Y I
+
[
1
.
We now derive a contradiction by using this to show that I must be precipitous. Suppose then that
[X] |- (
n
) [ n ) is an innite descending sequence in the generic ultrapower. For each n , let /
n
be a maximal antichain with respect to I such that whenever A /
n
there is an f
A
n
: V such that
[A] |-
n
=
f
A
n
. Let Y T(X) I
+
satisfy () for /
n
[ n ) and for each n let A
n
[ <
1
)
enumerate A /
n
[ A Y I
+
. Set B
n
= (A
n
<
A
n
) Y and B
n
=
<1
B
n
, and dene f
n
on B
n
by: f
n
=
<1
f
A
n
n
[B
n
S is an ideal over
1
which is
2
-saturated.
This result weakened the large cardinal hypotheses of previous results drawing the same conclusion, and
was an outgrowth of Foreman-Magidor-Shelah [FMS]. It was established using concepts and techniques of
the second author that we quickly review:
Suppose that P, ) is a poset for forcing, is regular with T(P) H
]
<1
consisting of N such that N, ) H
, countable N with N, ) H
, ) and N
1
S, and P-generic sequence p
n
[ n ) for N,
there is a q p
n
for every n . Semiproper is
1
-semiproper and S-closed is S-complete in the sense
of Shelah [Sh2].
1
-closure readily implies semiproperness. The salient features of these concepts are that if
a poset is semiproper, then forcing with it preserves stationary subsets of
1
(i.e. any stationary subset of
1
in the ground model remains stationary in the extension), and if it is S-closed, then it adjoins no new
countable sequences of ordinals.
To arm notation, P is the countable support iteration of P
,
Q
[ < ) i setting P
= P, we have:
P
0
= ; for < , |-
P
is a poset and P
+1
= P
[ < ) in case cf() > , and the inverse limit otherwise. Proceeding recursively we can take P
to consist of functions p (the conditions) with domain so that for each < , p() is a P
-name and
|-
P
p()
Q
1
Q
,
supt(p) = < [ |-
P
p() ,=
1
Q
p() q()).
Throughout the paper we rely on the following convention: For a notion of forcing P,
G
P
denotes the
6
3
9
0
r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
0
m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
1
canonical P-name for its generic object, and if P = P
is written
for
G
P
.
2.2 Lemma. Suppose that P is the countable support iteration of P
,
Q
[ ) such that:
(i) each N
[H
]
<1
, where = (2
1
)
+
, and S N
0
.
(ii) N
1
is an ordinal in
A [ A / N
.
Seal
S
(/) is ordered by: p q i q is an initial segment of p.
Seal
S
(/) is clearly S-closed, although it may not be semiproper, and forcing with it provides an enu-
meration of / in ordertype
1
and a closed unbounded subset of the diagonal union of / according to
that enumeration. If / was a maximal antichain with respect to NS, then it can be shown that forcing
with Seal
S
(/) preserves stationary subsets of
1
. / is then sealed: it remains a maximal antichain in
any extension that preserves stationary sets, since any stationary set has stationary intersection with the
diagonal union of /, and hence with a particular member of / by Fodors Lemma. The aforementioned
version of Martins Axiom implies through this means that NS[
S is
2
-saturated.
2.1 was established by applying reection properties directly in [Sh4: XVI]. Let
Q
S
= (
Seal
S
(/) [ Seal
S
(/) is semiproper) col(
1
, 2
1
) ,
the countable support product of all Seal
S
(/)s for /s that yield semiproper Seal
S
(/), followed by the
usual collapse of 2
1
to
1
using countable approximations. Clearly Q
S
is S-closed; it is also semiproper (see
[Sh4: XIII, 2.8(3)] or [Sh3: 2.8(3), p. 361]). Let P be the countable support iteration of P
,
Q
[ < )
7
3
9
0
r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
0
m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
1
where
Q
is a P
-name for Q
S
in the sense of V
P
. Assuming that satises the large cardinal hypothesis
of 2.1, the second author showed that any forcing extension via P satises its conclusion.
Proof of Theorem B. We interlace into the above described proof of 2.1 natural notions of forcing
for introducing suprema into T(
1
)/NS[
c [ w, c, d) G and d
G
=
d [ w, c, d) G.
Then c
G
is a closed unbounded subset of
1
, d
1
G
(0)A
S
is countable for every B B. In particular, if / B were a maximal antichain with respect to NS[
S, then
[d
1
G
(1)] would serve as an upper bound of / and [d
1
G
(0)] of B. These will be least upper bounds in the
extension if / B continues to be maximal there, and this is the only situation that will be germane to the
overall argument. Of course, for all this to make sense in the extension we must ascertain that stationary
subsets of
1
are preserved:
2.3 Lemma.
(a) Sup
S
(/, B) is
1
-closed (and hence semiproper and S-closed).
(b) Assuming CH, Sup
S
(/, B) is
1
-linked with least upper bounds, i.e. it is the union of
1
sets, each
consisting of pairwise compatible elements with least upper bounds.
Proof. (a) Set P = Sup
S
(/, B). Suppose that is regular with T(P) H
, N is countable with
N, ) H
, ) and p
n
[ n ) is a P-generic sequence for N. We must nd a q P such that q p
n
for every n .
For n and p
n
= w
n
, c
n
, d
n
), since w
n
is countable, there is a C
n
N be a closed unbounded subset
of
1
such that for any A w
n
/ and B w
n
B, C
n
AB
S = . Set w =
w
n
, c =
c
n
, d =
d
n
.
Then a simple genericity argument implies that c C
n
so that for no A w / and B w B does
c A B, and so we can nd an i < 2 such that w, c c, d c, i)) is in Sup
S
(/, B) and of course
is p
n
for every n .
8
3
9
0
r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
0
m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
1
(b) Note that w
0
, c, d) and w
1
, c, d) are compatible with least upper bound w
0
w
1
, c, d). With CH,
there are
1
such pairs.
Let
Sup
S
be the countable support product of Sup
S
(/, B) for all appropriate pairs /, B). A countable support product
of
1
-closed forcings is readily seen to be
1
-closed, and
1
-closed forcings are S-closed and semiproper.
Hence by 2.3(a), Sup
S
is S-closed and semiproper.
An approach to the proof of Theorem B would bec to carry out the countable support iteration of
Sup
S
through stages. Just assuming cf() >
1
, we would then get the consistency of 2
1
= and every
appropriate pair /, B) with [/ B[ < can be separated, a consequence in fact of a generalized Martins
Axiom in Baumgartner [B] or in Shelah [Sh1]. It is to ensure -saturation, a necessary condition for full
completeness by 1.1(a), that we build on the proof of 2.1.
Let P
,
Q
[ < ), where:
(i) For odd < ,
Q
is a P
-name for Q
S
in the sense of V
P
. (Here, Q
S
is as in the above outline
of the proof of 2.1.)
(ii) For even < ,
Q
is a P
[2 [ 2 < ]. (For Z ,
G
[Z is the P
-name for
p
G
[ supt(p) Z. Note that /, B) is to belong to the smaller model, but in the denition of appro-
priate pair the non-stationariness in (A /)(B B) A B
S is not stationary is to be in the sense
of V [G
]!)
Next, for set, by induction on :
P
= p P
name)
with the inherited order. (p() being a P
satises the conclusion of Theorem B. This follows from the following technical lemma, all
of whose parts are established by simultaneous induction; for its (b), note that P
-names being P
-names
is justied by an inductive appeal to (a), and Sup
S
was dened a few paragraphs ago in the outline of the
proof of 2.1.
2.4 Lemma. For each :
(a) P
< P
is a maximal antichain of P
.
(b) P
= p P
[ even supt(p)(p() is a P
is dense in P
.
9
3
9
0
r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
0
m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
1
(c) For any p, q P
such that p[2 [ 2 < and q[2 [ 2 < are compatible members of P
,
there is an automorphism F
pq
of P
such that: F
pq
(p) is compatible with q; F
pq
is the identity on P
; and
inductively for any < ,
F
p|,q|
= F
pq
[P
.
(d) If p P
.
Once this lemma is established, the proof of Theorem B can be completed as follows: Suppose that G
is P
= G
],
V [G
], so
that Sup
S
(/, B) at that stage would have adjoined a closed unbounded subset of
1
conrming that X is
non-stationary in V [G
S is -saturated.
It thus follows that NS[
S is also -saturated in V [G
].
We can conclude that for any maximal antichain / B of NS[
S in the sense of V [G
] with / B = ,
/, B) V [G
] for some even < . But then, the forcing with Sup
S
(/, B) would have adjoined a
set E
1
such that: if A /, then A (E
S) is not stationary; and if B B, then B E
S is not
stationary. [E
S] is thus the supremum for / in V [G
+1
] and hence in V [G
], T(
1
)/NS[
is S-closed, so that forcing with it adjoins no new countable sequences of ordinals. Conse-
quently,
p P
[ if supt(p) and w, c,
d) is a component of p(), then c = c and
d =
d for some c, d V
is dense in P
. Using CH and 2.3(b), a simple -system argument using this dense set then shows that P
has the
2
-c.c. and hence preserves all cardinals, and it is simple to see that it renders 2
1
= .
Proof of 2.4. Assuming that and all four parts hold below , we verify that they all hold at .
To rst verify that (b) holds at , suppose that p P
, ), p N, and N
1
S. Let p
n
[ n ) be a P
-generic
sequence for N with p
0
= p. Incorporating the proof of 2.3(a) into the iteration lemma for S-closed notions
of forcing, there is a least upper bound q P
for the p
n
s specied as follows:
supt(q) =
n
supt(p
n
). For even supt(q), q[ forces that for component w, c,
d) of q() in some
relevant Sup
S
(
/,
B) with corresponding w
n
, c
n
,
d
n
) in p
n
for n suciently large, w is the union of the
10
3
9
0
r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
0
m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
1
w
n
s, c is the union of the c
n
s together with its limit point at the top, and
d is the union of the
d
n
s together
with an arbitrary value for that top limit point. By P
-genericity of p
n
[ n ) and induction it can be
assumed that each w
n
is a P
d =
d for some d V .
Continuing to consider that specic component w, c,
d) of q(), by denition of such conditions there
is a P
-name w
0
such that q[ |- w w
0
, w w
0
) is an appropriate pair. By induction, P
< P
, and the
homogeneity property 2.4(c) implies that whenever r P
, is a P
( ). In particular
(q[)[2 [ 2 < |-
P
w w
0
, w w
0
) is an appropriate pair.
This conrms that q P
p()
Q
< P
, p
e
P
, and p
e
|-
P
p()
Q
. The
only problem in trying to replace P
by P
.
Assume to the contrary that for some q P
with q p
e
, q |-
P
p() /
Q
. By (b) inductively it
can be assumed that q P
, there is an automorphism F
q,p|b
as in (c) inductively such that F
q,p|
(q) is compatible with p[. Hence,
for some q
+
P
with q
+
q, F
q,p|
(q
+
) p[. Let G P
be P
= F
q,p|
G is P
. But G P
= G
as F
q,p|
is the identity on
P
, and so
/, B) [ / B T(
1
) V (G P
) = /, B) [ / B T(
1
) V [G
] ,
and so as V [G] = V [G
]),
/, B) V [G P
] [ /, B) is appropriate in V [G]) =
/, B) V [G P
] [ /, B) is appropriate in V [G
]) .
As also all -sequences from V of members of V [G P
] are in V [G P
)
G
= (
Q
)
G
. Setting q
e
= q[2 [ 2 < ,
q
e
P
p P
/
G
P
because if r q
e
P
then r
p,q
(r) is dened by: supt(F
p,q
(r))
i for some (equivalently, every) (, ) we have supt(F
p|,q|
(r)), and letting (F
p,q
(r))() =
(F
p|,q|
(r))() for some (any) such .
If = +1, odd, just note that
Q
is denable in V[
G
P
] there is an automorphism F
of
Q
[G] (note:
Q
[G] V [G P
]
see the proof of clause (d)) such that: F
(p()), and q() are compatible. (The simplest way to see this
is to replace in the iteration
Q
by its completion.)
3 Collapsing to
3
.
In this section we specify the modications to the proof of Theorem B necessary to establish Theorem
C. We dene the components of an iteration P
,
R
is dened as
Q
is dened as
Q
[Z is as before).
(iii) For 2 (mod 3),
R
is col(
2
, 2
2
), the collapse of 2
2
to
2
using
1
size approximations, in
the sense of V [
G
consist of functions p
with domain such that for each < , p() is a P
p()
R
, and
[supt(p) < [ 0 (mod 3) 1 (mod 3)[
0
[supt(p) < [ 2 (mod 3)[
1
.
The following lemma will be a consequence of forthcoming iteration lemmas.
3.1 Lemma. For each , P
= p P
name) .
(p() being a P
satises the
conclusion of Theorem C:
Let G
by P
-generic. It can be checked that the analogue of 2.4 holds in the new situation. In
particular, there is a G P
= G
], NS[
S is -saturated
and T(
1
)/NS[
for 2 (mod 3)
implies that in V [G
], = 2
1
3
. But by 3.7 below,
2
is preserved so that =
3
in V [G
] and so the
proof is complete.
The rest of this section is devoted to establishing 3.1 and the forthcoming 3.7. We build on Shelah
[Sh2][Sh4: XIV] and refer to them for the more basic details about iterated forcing that are not provided in
full.
For > , let /
be the class of Q,
Q
,
0
Q
) such that:
(i) Q,
Q
) is a semiproper, S-closed notion of forcing, say with maximum element
1
Q
.
(ii) Q,
0
Q
) is a poset so that: (a)
Q
renes
0
Q
, i.e. if p
0
Q
q, then p
Q
q; and (b)
0
is -closed,
i.e. if p
[ < ) is
0
Q
-decreasing and < , then there is a p Q such that p
0
Q
p
, we are referring
to the Q,
Q
) part.
Next, let /
be the class of P
,
Q
is a notion of forcing,
Q
is a
P
-name and |-
P
and recursively:
(i) P
, and setting
supt(p) = < [ |- p() ,=
1
Q
as before,
[supt(p)[
1
and
[ < [ (|- p()
0
1
Q
)[
0
.
(ii) The ordering on P
is given by:
p q i < (p[ |- p()
Q
q())
[ supt(p) [ (p[ |- p()
0
q())[
0
.
If P
is the iteration of P
,
Q
[ < ). The
introduction of the second partial order
0
serves less to provide iteration lemmas of potentially wide
applicability than to provide a uniform approach to 3.1. For that result,
0
R
will coincide with
R
when
13
3
9
0
r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
0
m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
1
2 (mod 3), i.e. when
R
R
will just be equality for 0 (mod
3) and 1 (mod 3). Note then that by how we dened the P
s from the
R
s, P
,
R
[ < ) /
.
The usual iteration facts hold for members of /
. For example, if P
,
Q
[ < ) /
, for any
< < , P
< P
is a maximal antichain of P
/P
such that P
= P
/P
, P
/P
is an iteration of a member of /
. To establish
3.1, we must verify that iterations in /
is the iteration of T = P
,
Q
[ < ) /
, ), N
1
S, T, , N, p
n
[ n ) is a
P
satises q p
n
[ for every n . Then there is a q
+
P
such that
q
+
[ = q and q p
n
for every n .
Proof. By adjusting names, we can assume for convenience in what follows that for each n and
< :
(i) p
n+1
[ |- p
n+1
()
0
p
n
() i |- p
n+1
()
0
p
n
(), and
(ii) |- p
n+1
() p
n
().
We now dene a function q
+
with domain as follows: Fix a well-ordering W of a suciently large V
.
Set q
+
[ = q. For < , having dened q
+
[ for < so that recursively q
+
[ P
, dene q
+
() as
follows:
(a) If for some k , q
+
[ |- p
n
() [ k < n < ) is
0
-descending in
Q
1
Q
, then since by
denitions of /
and /
, |-
P
0
Q
is -closed and > , there is a P
-name so that q
+
[ |- is a
0
-lower bound for p
n
() [ k < n < ). Let q
+
() be the W-least such . Else:
(b) If for some k and P
-name , q
+
[ |- p
n
() [ k < n < ) is
Q
-descending in
Q
1
Q
with a
Q
-lower bound, then let q
+
() be the W-least such . Otherwise:
(c) Set q
+
() =
1
Q
.
This denition perpetuates q
+
[ P
n
supt(p
n
[)[
1
.
Also, conditions on
0
and (i) above imply that there is a countable set E such that for E,
|- p
n+1
()
0
p
n
()
0
1
Q
, and so
< [ (|- q
+
()
0
1
Q
) E,
14
3
9
0
r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
0
m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
1
conrming that q
+
[ P
.
We next establish by induction on that
(1) for every n , q
+
[ |- q
+
() p
n
().
(2) If , N , then for every n , q
+
[ |- q
+
()
0
p
n
().
As N is countable, this suces to verify that q
+
p
n
for every n as desired. For < , the results
are immediate; what remains splits into two cases:
Case 1. < and N. Then P
,
Q
N, and we have q[ p
n
[ for every n by induction.
Since p
n
[ n ) is a P
-generic sequence
for N, and second that q
+
[ |- p
n
() [ n ) is a
Q
]. But by denitions of
/
and /
, |-
P
Q
1
Q
)
is countable, and clearly E
n
N, so that E
n
is countable in N. Hence,
n
E
n
N, so that ,
n
E
n
.
Hence, q
+
() was dened either through clause (a), or if not, neither through clause (b) as renes
0
,
but through clause (c). Both (1) and (2) now follow in this case also.
The following preservation result now follows in straightforward fashion:
3.3 Proposition. Suppose that P
is the iteration of P
,
Q
[ < ) /
/P
is S-closed.
The following lemma provides the main induction step for establishing the preservation of semiproper-
ness. Again, for a notion of forcing P,
G
P
denotes the canonical P-name for its generic object. For a set M
and q P, to say that q decides
G
P
M means of course that there is a set A V such that q |-
A =
G
P
M.
This happens for example if M is countable and q is a lower bound to a P-generic sequence for M, with
A = r P M [ q r. Finally, for > regular and N, ) H
, ) with P N, if G is P-generic
over V , then let N[G] be the set of interpretations [G] [ N is a P-name. By Shelah [Sh2: p.88],
() N[G], ) H
V [G]
, ).
We let N
[G
P
] be a canonical P-name for N[G].
3.4 Lemma. Suppose that P
is the iteration of T = P
,
Q
[ < ) /
, ), T, , N, N M, and
M
1
S. Assume also that q P
is N, P
)-semigeneric, q is M, P
M,
and p P
N with q p[.
Then there is a q
+
P
with q
+
p and q
+
[ = q such that q
+
is N, P
)-semigeneric, and q
+
is
M, P
and /
, |-
P
is
semiproper, and so with () just before 3.1 in mind, we have
() |-
P
r
Q
(r is N
[G
P
], P
M, by () applied syntactically there is a q
M
with q q
and a P
-name r
M such that q
. Since M
1
S,
|-
P
G
Q
M.
For a a limit, let
n
[ n ) enumerate M [ < with
0
= . Let
l
n
: n <
enumerate the P
), p
n+1
p
n
, and p
n+1
decides a value for
n
.
(b) q
n
is N, P
n
)-semigeneric, and q
n
is M, P
n
)-generic and decides
G
n
M.
(c) q
n
|-
n+1
= max
k
[ k n N
[G
n
].
(Note for (c) that if q
n
decides
G
n
M by (b), then N M implies that q
n
decides
G
n
N and hence
N
[G
n
].)
The case n = 0 follows from our initial assumption. Suppose now that q
n
has already been dened. By
() just before 3.4 applied syntactically, |- N
[G
n
] M
[G
n
] H
V [
Gn]
, )
. Moreover, since q
n
is M, P
n
)-
generic, we have q
n
|- M
[G
n
]
1
=
M
1
S. Hence, with
n+1
as stipulated by (c) it is straightforward
to apply the induction hypothesis in the sense of q
n
|- and then to nd an appropriate q
n+1
P
n+1
as
desired. There is no problem in dening p
n+1
.
We can now dene a q
+
P
so that supt(q
+
) =
n
supt(q
n
), and for any in this set, q
+
() = q
n
()
for some (any) n such that supt(q
n
). As in [Sh2][Sh4], q p
n
so q
+
is N, P
)-semigeneric. q
+
is not
necessarily (M, P
M, q
[ q
+
, q
is N, P
)-
semigeneric. Now we can nd a q
+
really as required.
16
3
9
0
r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
0
m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
1
The following preservation result is now clear, since for N, ) H
, ), and M
1
S by the stationariness of S.
3.5 Proposition. Suppose that P
is the iteration of P
,
Q
[ < ) /
/P
is semiproper.
Proof of 3.1. By 3.3 and 3.5 we get S-completeness and semiproperness.
Finally, we establish the preservation of
2
in a special case; a similar result appears in Shelah
[Sh2: VIII,1].
3.6 Proposition. Suppose that CH holds, and P is the iteration of T = P
,
Q
[ < ) /
2
where
for each < , |-
P
0
Q
=
Q
and Q
satises
2
-pic (see [Sh2: VII, Def.2.]) or |-
P
0
Q
is the
equality relation on
Q
()
0
p
, N
, and A
as follows:
Set p
0
= p. At limits <
2
with p
Q
s, let
p
()
0
p
().
To handle the successor stage, suppose that p
be countable with N
, ) H
, ),
T, , p
, and N
1
S. Then dene q
starting
with p
and then using the clauses (a),(b), and (c) as in the proof of 3.2. Hence q
is N
, P)-generic and
setting A
= r P N
[ q
r, we have q
|-
A
=
G
P
N
()
0
p
()
i q
[ |- q
()
0
p
() depends only on p
() and not on
(by carrying out the entire procedure canonically according to some well-ordering of a suciently large V
).
Finally, dene p
+1
as follows:
p
+1
() =
() if |- q
()
0
p
(),
p
() otherwise.
Proceeding with the proof, a straightforward -system argument with CH shows that there are
0
<
1
<
2
such that: N
0
1
= N
1
1
, and there is an isomorphism h: N
0
, , A
0
) N
1
, , A
1
) with
h[(N
0
N
1
) the identity, h(T) = T, h(
0
) =
1
, and h(p
0
) = p
1
. By assumption, there is an N
0
1
such that q
0
|- () =
0
. Hence, r A
0
(r |- () =
0
). Applying h, we have r A
1
(r |- () =
1
)
since h is the identity on N
0
1
. Consequently, q
1
|- () =
1
. However, it is not dicult to check that
q
0
and q
1
are compatible, reaching a contradiction:
17
3
9
0
r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
2
0
m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
1
-
2
1
The denitions of q
0
and q
1
as in the proof of 3.2 show that the countable sets
E
i
= supt(q
i
) [ (|- q
i
()
0
p
i
())
for i < 2 are such that E
i
N
i
and E
i
is countable in N
i
. We now argue that |- q
0
() and q
1
() are
compatible by cases, depending on whether E
0
, E
0
E
1
, or E
0
E
1
: If E
0
, then
|- q
1
() p
1
()
0
p
0+1
()
0
q
0
(). If E
0
E
1
, then |- q
0
() p
0
() and E
0
implies that
we are in the case |-
P
0
Q
is equality of our assumption about P. Consequently, , E
1
implies that
|- q
1
() = p
1
() = p
0
(), and so we have |- q
0
() q
1
(). Finally, if E
0
E
1
, then N
0
N
1
so the rst case in the proposition occurs and we apply
2
-pic. This completes the proof.
In 3.6 we can combine the two possibilities to one as implicit in the proof.
3.7 Proposition. P
,
Q
1
: < ) /
=
R
). Essentially, P
1
= P
preserve
2
. Now
the assumption of 3.6 clearly holds for P
1
,
Q
1
: < ) (for
2
-pic see the discussion of Q
S
for more).
This nally completes the proof of Theorem C.
Instead of 2
1
=
3
, for any regular such that
2
< , it is possible to arrange 2
1
=
+
=
using /