Professional Documents
Culture Documents
S. Shelah, C. Laflamme and B. Hart - Models With Second Order Properties V: A General Principle
S. Shelah, C. Laflamme and B. Hart - Models With Second Order Properties V: A General Principle
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
Models with second order properties V:
A general principle
S. Shelah C. Laamme B. Hart
October 6, 2003
Abstract. We present a general framework for carrying out the construc-
tions in [2-10] and others of the same type. The unifying factor is a com-
binatorial principle which we present in terms of a game in which the rst
player challenges the second player to carry out constructions which would
be much easier in a generic extension of the universe, and the second player
cheats with the aid of 3. 1 contains an axiomatic framework suitable for
the description of a number of related constructions, and the statement of
the main theorem 1.9 in terms of this framework. In 2 we illustrate the use
of our combinatorial principle. The proof of the main result is then carried
out in 3-5.
This paper is based on lectures given by the rst author while visiting the University
of Michigan. The research was partially supported by the BSF, the NSF and the NSERC.
1
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
Contents
1. Uniform partial orders.
We describe a class of partial orderings associated with attempts to man-
ufacture an object of size
+
from approximations of size less than . We also
introduce some related notions motivated by the forcing method. The un-
derlying idea is that a suciently generic lter on the given partial ordering
should give rise to the desired object of size
+
.
We describe a game for two players, in which the rst player imposes
genericity requirements on a construction, and the second player constructs
an object which meets the specied requirements. The main theorem (1.9) is
that under certain combinatorial conditions the second player has a winning
strategy for this game.
2. Illustrative application.
We illustrate the content of our general principle with an example. We
show the completeness of the logic L
<
, dened by Magidor and Malitz [2]
for the
+
-interpretation assuming the combinatorial principles Dl
and 3
+.
3. Commitments.
We give a preliminary sketch of the proof of Theorem 1.9. We then
introduce the notion of basic data which is a collection of combinatorial
objects derived from Dl
<
A
.
(Note the exception arising in lemma 1.3.)
Throughout the paper, is a cardinal such that
<
= .
T
<
(A) = B A : [B[ < .
otp (u) will mean the order type of u. Trees are well-founded, and if T is
a tree, T, we write len () for otp T : < (the level at which
occurs in T).
Acknowledgements
We would like to express our deepest gratitude to Greg Cherlin without
whose insistence this paper might not have been completed this century.
3
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
1 Uniform partial orders
We will present an axiomatic framework for the construction of objects of
size
+
from approximations of size less than , under suitable set theoret-
ical hypotheses. The basic idea is that we are constructing objects which
can fairly easily be forced to exist in a generic extension, and we replace
the forcing construction by the explicit construction of a suciently generic
object in the ground model.
We begin with the description of the class of partial orderings to which
our methods apply. Our idea is that an approximation to the desired nal
object is built from a set of ordinals u
+
of size less than . Furthermore,
though there will be many such sets u, there will be at most constructions
applicable to an arbitrary set u. We do not axiomatize the notion of a
construction in any detail; we merely assume that the approximations can
be coded by pairs (, u), where < is to be thought of as a code for the
particular construction applied to u. An additional feature, suggested by
the intuition just described, is captured in the indiscernibility condition
below, which is a critical feature of the situation though trivially true in
any foreseeable application.
Denition 1.1 A standard
+
-uniform partial order is a partial order
dened on a subset P of T
<
(
+
) satisfying the following conditions,
where for p = (, u) in P we write domp = u, and call u the domain of p.
1. If p q then domp domq.
2. For all p, q, r P with p, q r there is r
P so that p, q r
r and
domr
= domp domq.
3. If (p
i
)
i<
is an increasing sequence of length less than , then it has a
least upper bound q, with domain
i<
domp
i
; we will write q =
i<
p
i
,
or more succinctly: q = p
<
.
4. For all p P and <
+
there exists a q P with q p and
domq = domp ; furthermore, there is a unique maximal such q, for
which we write q = p.
5. For limit ordinals , p =
<
p.
4
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
6. If (p
i
)
i<
is an increasing sequence of length less than , then
(
i<
p
i
) =
i<
(p
i
).
7. (Indiscernibility) If p = (, v) P and h : v v
+
is an order-
isomorphism then (, v
P for so that
5
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
1. dom(r
) and
2. r
p, q.
Let r
.
If dom(q) dom(p) then p + 1 = p. Apply weak amalgamation
to r
.
Since these are all the possibilities, let = sup(dom(p) dom(q)) and
so r
= p P : domp
for <
+
. Note that P
+ = P.
Be forewarned that the following denition does not follow the standard
set theoretic use of the term ideal.
Denition 1.2 1. For <
+
, a -generic ideal G in P
is a subset of
P
satisfying:
(a) G is closed downward;
(b) if Q G and [Q[ < then Q has an upper bound in G; and
(c) for every p P
.
6
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
2. If G Gen(P
) then
P/G = p P : p is compatible with every r G.
Note that p P/G i p G.
3. We say an increasing sequence g
i
: i < ) is conal in G Gen(P
) if
G = r P
: for some i, r g
i
. Every G Gen(P
) has a conal
sequence of length (possibly constant in degenerate cases). We will
often write (g
to mean g
: < ).
4. We will say that G is generic if G Gen(P
) for some .
Lemma 1.3 Let G
i
Gen(P
i
) for i < be an increasing sequence of sets,
and = sup
i
i
. Then there is a unique minimal -generic ideal of P
containing
i<
G
i
. This ideal will be denoted G
<
.
Proof: We may suppose that is a regular cardinal, . If = then
it is clear that
i<
G
i
Gen(P
)
<
conal in G
i
. Fix i < j < . There
is a club C
ij
in such that for all C
ij
, g
i
= g
j
i
. Let C =
i<j<
C
ij
.
If C then dene g
=
i<
g
i
. 2
The notion of -genericity is of course very weak. In order to get a notion
adequate for the applications, we need to formalize the notion of a uniform
family of dense sets.
Denition 1.4
1. For <
+
and G Gen(P
T
<
(
+
),
8
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
if there is an order-isomorphism h : w
w with h[u
] = u, then let
D
p,q,
(u
, w
) = r : domr w
, w
) = r : domp w
. We claim
that D
p,q,
is a 0-density system. It suces to check the density condition
for u w , and this is immediate.
Application
We will now show that if G Gen(P
, h[u]) where
= h
0
[p], q
= h
0
[q],
= h
0
(),
and u = domq. Now choose w with [w[ < and r GD
p
,q
,
(u, w).
Since X
with h[w] = w
.
We claim that there is s p
, q
p , q , r in G. Since
q P/G and r
with
dom q = domq domr
,q
,
(u, w), it now follows that there exists s p
, q
so that
s
G with
9
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
G
Gen(P
, either q G
. We will
nd another family of at most 0-density systems D
p,q,r,
which make it
possible to construct a suitable G
G if G meets all D
p,q,
(from Example
1.5) and D
p,q,r,
.
Construction
For p, q, r P
T
<
(
+
), if there is an
order-isomorphism h : w
] = u then let
D
p,q,r,
(u
, w
) = s : doms w
, w
) = s : doms w
.
We claim that D
p,q,r,
is a 0-density system. Again we check only the
density condition for u w . So suppose we have s P, doms w, and
s is compatible with r. We seek s
s in D
p,q,r,
(u, w).
Choose s
. Then s
r p and doms
, p so that
dom(t) = doms
p, q, r). Now t s
r, s, so if s
= t then s
r, s, and
s
D
p,q,r,
(u
, w
) as desired.
Application
We return to the situation in which we have G Gen(P
), > , and
we want to build G
G with G
Gen(P
is xed. We wish to
10
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
decide q: that is, we seek p p so that either p is incompatible with q, or
else p q.
If p is already incompatible with q then let p = p. Otherwise, let X =
domp domq and let h : X X
= h[p], q
= h[q], and
,q
,
(u, w). Extend h to an order-preserving function from X w
to X
, and let r
= h[r].
Suppose rst that there is some s p
, q
with s
. We may suppose
that doms = domp
domq
, q
, r
. We may assume
that p r since p G and G is directed. Let:
Y = domp domq domr ,
and let k : Y Y
be the -collapse of Y . Let p
= k(p), q
= k(q), r
=
k(r), and
= k(). Then p
, and there is no s p
, q
, r
.
Let v = (domqdomr), and choose z v and s GD
p
,q
,r
,
(v, z).
We can extend k to an order-isomorphism from Y z to Y
z
with k[z] =
z
. Let s
= k[s].
Certainly r
and s
,q
,r
,
(u, z), we have k[s] r
so that
t
and t
is incompatible with q
<
+
, and will have dened an increasing sequence of -
generic ideals G
on P
P/G
<
and will also choose at most density systems D
i
over
G
<
. Note that G
<
Gen(P
<
) by Lemma 1.3.
11
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
2. After player I has played his
th
move, player II will pick an ordinal
.
Player II wins the P-game if the sequences
and G
meets D
i
.
Our main theorem uses the following combinatorial principle.
Denition 1.8 Suppose is a regular cardinal. Dl
T(), [/
is stationary.
Easily, 3
. Also,
Kunen showed that Dl
+ implies 3
implies
<
= .
Theorem 1.9 Dl
1
. They ask whether these axioms
are complete for the
+
-interpretation. We will show that their axioms are
complete when both Dl
and 3
+( <
+
: cf() = ) hold. This will
explain a remark at the end of [5]. See Hodges ([1]) for a treatment in the
same vein for the
1
-interpretation.
Fix a complete L
<
theory T, [T[ . Let Q = Q
1
. We may assume that
that associated to each formula with free variables x
1
, . . . , x
n
, y
1
, . . . , y
m
,
L contains an (m+1)-ary function F
(, y
1
, . . . , y
m
) is one-to-one and
Q
n
x(x
1
, . . . , x
n
, y
1
, . . . , y
m
)
z
1
. . . z
n
(
i<j
z
i
,= z
j
Sym(n)
(F
(z
(1)
, y), . . . , F
(z
(n)
, y), y)).
Strictly, it is not necessary to make this conservative extension to our lan-
guage and theory but it is convenient when handling the inductive step cor-
responding to Q
n
.
We add new constants y
: <
+
and x
i
: <
+
, i < to L,
to obtain a language L
1
. The set of constants y
i
: i < is called
the set of -constants and y
be
p P : all constants occurring in a formula of p are (< )-constants.
The elements of P
, w T
<
(
+
) where otp(w) =
p
by sending
([p], w) to p(w). Throughout the rest of this section we will treat P as if
it were in standard for although in practice we will use its original deni-
tion. We claim that P is
+
-uniform; it suces to check the amalgamation
condition 8.
The following notation will be convenient. If (y
1
, z
1
, y
2
, z
2
, . . . , y
n
, z
n
)
is a formula with
1
>
2
> . . . >
n
, and z
i
is a collection of [
i+1
,
i
)-
constants, then the string S of quantiers:
x
n
Qy
n
. . . x
1
Qy
1
is called standard for where the xs quantify over the zs. Its dual is denoted
S
:
x
n
Qy
n
. . . x
1
Qy
1
and p q.
Suppose ( x, y
and z is
the < -variables. Then
Qy x( x, y, z) p
.
If q then
SQy x( )
where S is a standard sequence, is equivalent to
S( Qy x).
Since both of the conjuncts are in q, this last sentence is in T. This veries
weak amalgamation.
Now the strategy is to build a set G which is the union of generics so
that the constant structure derived from G will form a model of T under the
+
-interpretation. More precisely, we introduce an equivalence relation on
the set of nonspecial constants A = x
j
: <
+
, j < by:
a b i a = b G.
Let
G = a/: a A and dene the functions and relations on
G in the
usual manner. We want to ensure that for any formula in L
<
1
we will
have:
G [= (a
1
/, . . . , a
n
/) i (a
1
, . . . , a
n
) G. (1)
If (1) is true, its proof naturally proceeds by induction on the complexity
of formulas. We now describe a strategy for Player I in the genericity game
which can only be defeated by achieving (1). In other words, we will specify
density systems and elements g P, to be played by Player I, such that a
proper response by Player II ensures that G allows an inductive argument
for (1) to be carried out. Our discussion will be somewhat informal, stopping
well short of actually writing down the density systems in many cases.
15
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
We begin with the treatment of the ordinary existential quantier. When-
ever x(x, z) G we will want (eventually) to have some a A so that
(a, z) G. In particular, for every there will be some a A so that
y
from time to time. We will also have to deal with the case in which
Qx(x), and we will return to this in a moment.
We now consider the quantier Q
n
. Suppose that the formula Q
n
x( x, y)
is in G at some stage. This is where we use the function F
. If
G is a model
of T then it has cardinality
+
. Moreover, F
(, y) is one-to-one. Since
Q
n
x( x, y) is in G, so is
z
1
. . . z
n
(
i<j
z
i
,= z
j
Sym(n)
(F
(z
(1)
, y), . . . , F
(z
(n)
, y), y)).
It follows that the range of F
(, y) is homogeneous for .
We are now left with the cases in which formulas of the form Q
n
x
(n 1) are placed in G. We deal rst with the case n = 1. For this case, we
dene a number of density systems depending on the following parameters:
1. j, j
0
, . . . , j
m1
< ;
2. a formula (x, y
0
, . . . , y
m1
); and
3. a function f : m m.
We associate with these data a density system D. If otp (u) ,= m+1, we
let D(u, w) be degenerate:
D(u, w) = p P : dom() .
If otp(u) = m + 1 then let g : m + 1 u be an order preserving map, let
h = gf and set = g(m) = max u, and:
(x) = (x, x
h(0)
j
0
, . . . , x
h(m1)
j
m1
).
We will then let D(u, w) consist of those p P for which, setting = min(w),
we have:
16
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
1. dom(p) w;
2. If Qx(x) p, then either (x
j
) p or x
j
= x
i
p for some i < .
We shall verify the density condition on D. Suppose q P and
Qx(x) q. The extension of q we are about to construct will only involve
the adjunction of formulas with ( )-constants, so we may assume that q
itself contains only ( )-constants.
If we cannot complete q(x
j
) to an element of P, then there is some
q so that:
Sx( (x)) , T
where Sx is a standard sequence for the formula . Note that by the
assumption that is the maximal element of dom(q), we may assume that
the nal quantier in the standard sequence is an existential quantier on
the constant x in .
By the axioms for the Q-quantier, for any q such that T ,
Sx( (x)) T.
As Qx(x) q, repeated use of the Q-quantier axiom:
Qxy(x, y) xQy(x, y) Qyx(x, y)
shows that xS( (x)) T.
If we now choose a constant x
i
not occurring in q, where = min(w),
one can conclude that q x
j
= x
i
can be completed to an element of P.
It is easy to see that if the foregoing density systems are met, then we
can carry out the argument from right to left in condition (1) above for
= Qx(x). We turn now to the treatment of the quantiers Q
n
for n > 1.
By applying Fodors Lemma to the map sending to dom(f()) we
obtain:
Lemma 2.1 If S : cf () = is stationary and f : S P then
there is a stationary S
,
f() = p(w
) where w
= dom(f()) and w
.
It will be convenient to treat conditions as if they were single formu-
las. Extending our previous notation, for p P and S a standard sequence
covering some of the variables in p, we will write S(p) for the set:
S
: p
17
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
where S
)
cf ()=
be a 3-sequence. For u, v sets of ordinals, we write u < v
if for all u, < min v.
If cf () = and G
Gen(P
1jk
w
j
;
2. < < w
k
< . . . < w
1
;
3. w
j
=
p
j
;
4.
1jk
p
j
(w
j
) can be extended to a member of P;
in which case we adopt the following notation. Let
j
be the
th
j
element of
w
j
, and write z
for x
i
. Note that since
j
<
p
j
we will have
j
> and
hence <
k
< . . . <
1
. Dene the set r(
1
, . . . ,
nk1
) for
1
< . . . <
nk1
A
to be
S(
1jk
p
j
(w
j
) (z
1
, . . . , z
nk1
, z
, z
k
, . . . , z
1
))
where S covers all the (> )-variables.
We now dene D(u, w) as the set of q P/G with dom(q) w which
satisfy one of the following three conditions:
18
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
1. q
1jk
p
j
(w
j
); or
2.
1jk
p
j
(w
j
) q and for some
1
< . . . <
nk1
A
,
r(
1
, . . . ,
nk1
) q; or
3.
1jk
p
j
(w
j
) q and for all
1
< . . . <
nk1
A
:
S
0
(q S
k
(p
k
(w
k
) . . . S
1
(p
1
(w
1
)
(z
1
, . . . , z
nk1
, z
, z
k
, . . . , z
1
)) . . .)) G
, there
is a q and
j
p
j
(w
j
) so that
S
0
( S
k
(
k
. . . S
1
(
1
(z
1
, . . . , z
nk1
, z
, z
k
, . . . , z
1
)) . . .)) G
where S
j
covers all the ( )-variables in
j
for j > 0, and S
0
covers all the
( )-variables in . Notice that the only overlap among the variables occur
in the (< )-variables.
Now suppose
G [= Q
n
x( x, a/). We would like to argue that Q
n
x( x, a)
G. For convenience we will suppress the parameters a. We may also assume
that T ( x)
i<j
x
i
,= x
j
.
Since
G [= Q
n
x( x), there is a
+
-homogeneous subset B
G for . We
may assume there is an i < so that every b B is of the form x
i
/ for
some . Let A = : x
i
/ B and is the least such in a given -class .
For any so that cf () = , let
= min(A A
). Note that if A = A
then
> .
We will now produce the following data. There will be:
1. stationary sets S
k
for 0 k n with S
k+1
S
k
for all k < n and
S
0
= : cf () = , and A = A
;
2. templates p
k
for 1 k n, and ordinals
k
so that
k
<
p
k
;
3. a domain w
k
k
; let
k
be the
th
k
element of w
k
;
19
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
4. For 0 k < n, if <
k
< . . . <
1
S
k
are chosen so that
k+1
< w
k
k
k
< . . . < w
1
1
1
and D is the density system over G
corresponding to i, , k, p
1
, . . . , p
k
and
1
, . . . ,
k
, then
p
k+1
(w
k+1
) D(
k+1
w
k
k
. . . w
1
1
, w
k+1
) G
Using lemma 2.1 and the fact that G meets all the density systems intro-
duced at stages S
0
, this is straightforward.
Now suppose
n
< . . . <
1
S
n
, so that w
n
n
n
< . . . < w
1
1
1
. Let
q
k
= p
k
(w
k
k
).
Since B is a homogeneous set for , it follows that (z
n
, . . . , z
2
, z
1
)
G, for every
n
< . . . <
2
<
1
in A
. Since q
1
G, using the density
systems dened before, we conclude that
S
1
(q
1
(z
n
, . . . , z
1
)) G
where S
1
covers the w
1
1
-variables. Proceeding by induction and using the
denition of the density systems, we conclude that
S
n
(q
n
S
n1
(q
n1
. . . S
1
(q
1
(z
n
, . . . , z
1
)))) G
where S
j
covers the w
j
j
-variables.
Of course, S
n
q
n
G, so by the Magidor-Malitz axioms, Q
n
x( x) G
and we nish.
20
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
3 Commitments
In this section we begin the proof of Theorem 1.9. Our main goal at present
is to formulate a precise notion of a commitment (that is, a commitment
to enter a dense set or in model theoretic terms, to omit a type). We will
also formulate the main properties of these commitments, to be proved in
4-5, and we show how to derive Theorem 1.9 from these facts.
Before getting into the details, we give an outline of the proof of Theorem
1.9.
General overview
Suppose that we wish to meet only the following very simple constraints.
We are given some 0-density systems D
i
over for i < , and some g
0
P,
and we seek a -generic ideal G
0
containing g
0
, and meeting each D
i
. Let
= sup(dom(g
0
)), and enumerate P
as (r
i
: i < ). Then we may
construct G
0
by generating an increasing sequence (g
)
<
beginning with
the specied g
0
, and taking G
0
to be the downward closure of (g
). We will
run through this in some detail.
Our rst obligation is to make G
0
-generic in P
so that either r g
or else
r g
. If the sequence (g
)
<
ultimately decides every r P
, then G
0
will
be -generic in P
is decided. This
takes care of the basic -genericity requirement. At limit stages we can let
g
g
<
with g
D
i
(u, v).
Thus it is easy to deal with constraints of the type arising in one play
of our genericity game. Our strategy in that game will rely on this sort
of straightforward do what you must when you have the time approach,
but will encounter diculties in keeping up at limit stages in the game.
We will use Dl
.
At stage 0, player I selects some density systems, to which we may add
all the density systems from examples 1.5 and 1.6. From these we construct
some stage 0 commitments
0
p, and a G
0
meeting
0
p.
At stage in the play of the game, Player II is attempting to extend
G
<
to a suitable G
meeting
p as above. In this way, Player II wins the game.
There is a certain diculty involved in coping with the freedom enjoyed
by Player I (in terms of obligations accumulating at limit stages in the game).
There are a priori
+
sets u T
<
(
+
) that may require attention. On the
other hand, at a given stage we are only prepared to consider fewer than
such sets. However, by uniformity, it will be sucient to consider pairs
(u, w) T
<
( + ) T
<
( +), and hence such pairs suce. This still
leaves Player II at a disadvantage, but with the aid of Dl
, at limit stages we
will guess a relevant set of us of size less than .
It remains to show that this strategy can be implemented, and works.
We introduce the notion of basic data which will be provided by Dl
.
Denition 3.1 A collection of basic data will contain
1. trees T
, subsets of P
, for every
< ;
2. for every generic set G Gen(P
(G) S(G);
and
3. for every < , a set U
T
<
()
with the following properties
1. [T
[ < , [U
,
5. if (g
) then there
is a club C so that for C S(G), (g
<
)
col
T
,
6. if G and G
) S(G) and
7. (oracle property) for <
+
and G Gen(P
), u T
<
() and
=
<
w
T
<
() and
u w
0
then there is a club C so that for every C S
(G) there is
u
so that (w
, u)
= (otp (w
), u
).
Remarks:
1. Although there is the possibility of confusion between the orders <
and < on P
, we will use < for both and the context should usually
make it clear which we mean.
2. The following will be true of the trees that we eventually build although
this property will not be needed in the proof: if q T
then there is
a generic set G with a conal sequence (g
so that q = (g
col
<
) for
some and
3. If (g
and (g
then there is a
club C so that if C and = dom(g
col
<
) then g
col
<
= (g
<
)
col
. In
condition 6, we may assume that for particular conal sequences, C
satises this property as well as C S(G
: < ) where
p
: T
() P
len ()
(We usually write p
for p
().)
2. if T
then p
len ().
We dene an order on weak commitments by p q if for almost all
(i.e. on a club), p
: < ) where
p
= .
Denition 3.4 A commitment is a weak commitment which is stronger than
p.
Denition 3.5 Suppose G is generic with a conal sequence (g
and p
is a commitment. We say G meets p if there is a club C in so that
for every C S(G),
=: (g
<
)
col
T
and there is r
G so that
dom(r
) = dom(g
<
) and
r
col
for all
.
Remark: If h : len (
) dom(g
<
) is an order isomorphism then in the
above denition the existence of r
] G for
all
.
Proposition 3.6 If D
i
, i < , are 0-density systems and g P
then there
is a commitment q(
p) and some G Gen(P
), so that:
1. g G,
2. G meets q and
3. if
<
+
and G
Gen(P
) meets q, then G
meets each D
i
.
24
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
Proposition 3.7 Suppose G Gen(P
) and G satises
1. for all g P/G, P there is
P/G with
and either
or
and
2. P/G is
+
-uniform.
For i < , let D
i
be a density system over G, and suppose g P
/G where
<
+
and p is some commitment that is met by G. Then there is a
commitment q p, and some G
Gen(P
), so that:
1. G G
, g G
;
2. G
meets q;
3. If
<
+
and G
Gen(P
meets each D
i
.
Lemma 3.8 Let (
p)
<
be an increasing sequence of commitments with
<
+
. Then the sequence has a least upper bound.
Notation 3.9
With the notation of the preceding lemma, we write
<
p or
<
p
for the least upper bound of the commitments
p.
Proposition 3.10 Suppose G
Gen(P
) meets
p for all < , where
<
+
, and the G
and
p are increasing. Then G
<
meets
<
p.
By combining these results we immediately obtain a proof of Theorem
1.9.
Proof of Theorem 1.9: We dene a sequence of ordinals
, a sequence of
commitments
p, and a sequence of -generic ideals G
Gen(P
), so that:
1.
<
;
<
p
p; G
<
G
for <
+
2. G
and G Gen(P
) contains G
proposed by Player
I at stage of the genericity game.
At stage 0, Player I provides some g
0
P and at most many 0-density
systems. To these Player II adds all the 0-density systems mentioned in
examples 1.5 and 1.6. We now apply Proposition 3.6 to all these 0-density
systems and g
0
. This will provide us with G
0
,
0
and
0
p.
At stage , we will have
<
, G
<
,
<
p dened, and by Proposition 3.10
G
<
meets
<
p. Now since G
0
G
<
, G
<
meets
0
p and hence meets each of
the 0-density systems from examples 1.5 and 1.6. It follows that G
<
satises
the conditions on the generic set in Proposition 3.7. By Proposition 3.7 a
suitable choice of
, G
,
p can then be made.
Now we verify that Player II wins the genericity game using this strategy.
By construction, g
, it follows that G
meets D. 2
26
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
4 Proofs
In this section we give the proofs of the results stated in the previous section
except the proofs of Propositions 3.6 and 3.7 which are deferred to the next
section.
Proof of Lemma 3.2
The Lemma states that there is a collection of basic data.
Let (v
)
<
be an enumeration of T
<
() so that v
= v
and an encoding of
( ) ( T
<
())
by , we can nd sets 1
T( ) and (
T( V
), such that
[1
[, [(
and G ( V
) (
is stationary.
Before dening the basic data we establish some notation. For each <
+
, we select a bijection i
: i
1
[]
, j
i
1
.
Let
R
= (i
(), i
= R
( ) ( < ).
Then j
: (, R
) (
is determined by
R
.
If G is a -generic ideal in P
with <
+
, let:
G = (, i
[u]) : (, u) G
=
G ( V
)
G
= (, u) G : (, i
[u]) ( V
= (,
[u]) : (, u) G
27
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
Again, we can go directly from
G
to
G
by applying j
. Observe also
that
induces an isomorphism
: G
, but
G provides a better encoding of G because
the sets
G
do not.
Let C(G) be the set of < for which G
.
We are now ready to dene the basic data. For < we dene T
as:
p P : <
+
G Gen(P
) : C(G),
, R
, and p = [
].
Notice that dom(p) is an ordinal for every p T
. To see that [T
[ < , we
use the fact that
G
, R
together determine
G
with <
+
, x a conal sequence
(g
G
i
)
i<
in G, and set:
S(G) = < : [g
G
<
]
col
T
;
S
(G) = < :
G
and R
;
U
= u : v V
R 1
<
+
(, v, R) (, u, <)
Clearly U
T
<
() and [U
(G) is stationary.
Let C
1
be
C(G) : g
G
<
= G
Then C
1
is a club in , and if S
(G) C
1
then (g
G
<
)
col
T
so S(G) is
stationary. If (g
<
= g
G
<
<
)
col
T
.
Let G G
, P
) de-
termined by conal sequences (g
G
, (g
G
<
= g
G
<
, it is easy to see that C S(G
) S(G).
28
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
It remains to verify the oracle property (7) of Denition 3.1. We x
<
+
, G -generic in P
, u T
<
(), and we let =
<
w
be a
continuous increasing union with each [w
[ < and u w
0
. On some club
C, otp w
and if v
then < . So (w
, u) (
[u]). For
C S
(G) we have
(, i
[u], R
) (
[u], <) (w
, u, <).
Hence,
[u] U
. 2
Notation 4.1 In the next few results we make systematic use of the diag-
onal intersection of clubs. If (C
)
<
is a sequence of clubs in , the diagonal
intersection is dened correspondingly as:
= < :
<
C
.
The diagonal intersection of such a sequence of clubs is again a club.
Proof of Lemma 3.8 Let (
p)
<
be an increasing sequence of commit-
ments with <
+
. We claim that the sequence has a least upper bound.
We may take to be a regular cardinal, with . We deal with the case
= ; for < our use of a diagonal intersection below would reduce to an
ordinary intersection.
For < let C
pointwise for C
.
Let C =
. For C and T
, let p
=
<
. Then p is a
commitment. We have
p p since
p
pointwise for C .
Now we will check that p is the least upper bound of the sequence as a
commitment Let q be a second upper bound. Let
C
= < : q
pointwise,
and let C
. For C C
, and T
, we have:
q
<
= p
.
It follows that p is the least upper bound. 2
We divide Proposition 3.10 into two parts.
29
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
Proposition 4.2 Suppose that (G
i
)
i<
is an increasing sequence with G
i
Gen(P
i
), <
+
, and that each G
i
meets a xed commitment p. Then
i
G
i
also meets the commitment p.
Proof: Let G =
i<
G
i
. By consulting the proof of Lemma 1.3, we
can see that there is a conal sequence (g
j
)
j<
for G so that if g
i
j
= g
j
i
then (g
i
j
)
j<
is a conal sequence in G
i
. For each i < let C
i
be a club
demonstrating that G
i
meets p. In other words, for C
i
S(G
i
) we have
r
i
G with:
1. domr
i
= domg
i
<
,
i
= (g
i
<
)
col
T
; and
2. [r
i
]
col
p
for all
i
.
By Denition 3.1.5, we may also suppose that C
i
S(G) S(G
i
).
We consider the case = (Use ordinary intersection instead of diagonal
intersection when < ).
Let
C =
i
C
i
< : sup domg
<
= sup
i<
G so that r
r
i
) =
i<
dom(g
i
<
) =
i<
dom(g
<
i
) = dom(g
<
).
Clearly, r
col
. However, since p
len ()
p
for all
and p
<
len(), r
col
. 2
Proposition 4.3 Suppose G Gen(P
p.
Proof: Again, we treat only the case when = . Let (g
i
)
i<
be a
conal sequence in G. For each , let C
S(G) there is r
G so that:
1. domr
= domg
<
,
= g
col
<
T
; and
2. (r
)
col
for all
.
30
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
We may also suppose that C
for C
S(G). Let C =
.
For C S(G), let r
i<
r
i
= domg
<
and r
col
=
i<
[r
i
]
col
i<
i
p
=
<
p
for all
) and G satises
1. for all g P/G, P there is
P/G with
and either
or
and
2. P/G is
+
-uniform.
For i < , let D
i
be a density system over G, and suppose g P
/G where
<
+
and p is some commitment that is met by G. Then there is a
commitment q p, and some G
Gen(P
), so that:
1. G G
, g G
;
2. G
meets q;
3. If
<
+
and G
Gen(P
meets each D
i
.
We are also obliged to prove Proposition 3.6 as well. The proof is very
similar to the proof of Proposition 3.7 and so we will only highlight the formal
dierences at the end of the proof.
Proof of Proposition 3.7: Let =
<
w
be a continuous increasing
union with [w
[ < . Set
= otp(w
) and choose
so that
ht (T
).
Let
h
[, +
)
be an order isomorphism.
Fix a conal sequence (g
, and there is r
G so that
dom(r
) = w
and r
col
for all
for all C.
Now we build the commitment q. If , C S(G) then let q
= p
. Fix
then C S(G). For i < ,
and u U
, u
, let
D
i
(u) = r P
+
: h
[r] D
i
(h
[u], w
[, +))
Let P
G
[T
so that
32
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
1. p() P
len ()
for all T
,
2. if then p() p()len () and
3. if is comparable with
then h
[ p()] P/G.
We will write p
for p().
Remark: Since G meets p, if C S(G) then p
P
G
[T
]. To see this,
we must show that if is comparable to
then h
[p
] P/G. If
then
since C S(G), h
[p
] G. Suppose
. Now p
len (
) p
and
since w
= dom(g
<
) we have h
[p
] h
[p
] so h
[p
] P/G.
Proposition 5.1 There is a q
P
G
[T
] with q
, i < if
with len (
) =
+ and u
+
then q
i
(u).
To obtain this q
], u U
, i < ,
and
with
len (
) =
+ and u
+ then there is r P
G
[T
] so that r q
pointwise and r
i
(u).
Proof of Proposition 5.1 To get the required q
, at
limit stages take unions and at successor stages use the claim applied to some
particular i < ,
, u U
and
. After at most [U
[ [[ [
[ [T
[
stages we will have produced q
. 2
Now we turn to the construction of G
, a -generic ideal in P
meeting q
with G G
and g G
of P
. G
0
= g. We shall guarantee that g
/G for each .
At stage , if C S(G), dom(g
<
) = w
, (g
<
) = g
<
and (g
<
)
col
=
) dom(g
<
) be an order isomorphism and let
g
) = dom(g
<
), g
<
and g
[q
]
for every
[q
= g
<
. In either case, use the
assumption on G to nd g
so that g
/G with g
<
and either
g
or g
.
It follows easily now that G
Gen(P
), G G
and g G
. We now
show that G
meets q. Choose C
1
so that C
1
S(G
<
) = w
and (g
<
) = g
<
. If C
1
S(G
) then
(g
<
)
col
=
and ( g
)
col
q
for all
. It follows that G
meets q.
Now suppose G
meets q, G
Gen(P
) contains G with
<
+
.
Fix a conal sequence (g
for G
, a density system D
i
and u T
<
(
).
We want to nd w so that u w and D
i
(u, w) G
,= . Write
as a
continuous increasing union
<
w
with w
T
<
(
).
There is a club C
2
with the following properties:
1. if C
2
, (g
<
) = g
<
and dom(g
<
) = w
;
2. C
2
S
(G
) C
2
S(G
) C S(G);
3. for C
2
S(G
) there is r
with dom(g
<
) = dom(r
), g
<
r
,
(g
<
)
col
=
and r
col
for all
; and
4. for C
2
S
(G
) there is u
so that
f
: (otp(w
), u
) (w
, u).
This can be obtained by refering to the denition of basic data, Deni-
tion 3.1 and Lemma 3.2. In particular, condition 4 follows from the oracle
property.
Now choose C
2
S
(G
). Since
dom(g
<
) = w
and
= (g
<
)
col
T
, we have that
+ ht (T
).
Moreover,
= g
col
<
T
and
. Hence,
h
[q
] D
i
(h
[u
], w
[, +))
and r
col
with r
.
By the indiscernibility of the density systems, we have
f
[q
] D
i
(u, w
)
34
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
since
f
h
1
: (w
[, +), h
[u
]) (w
, u).
By the indiscernibility of P, we have r
[q
]. Since r
,
f
[q
] G
D
i
(u, w
)
so G meets D
i
. 2
It remains to prove Claim 5.2.
Proof of Claim 5.2: Consider the set
S = h
[ q
] :
P
len (
)
so that
r
for all
and h
[r
i
(u) so that r
len() if
otherwise.
It is easy to check that r P
G
[T
]. 2
To obtain a proof of 3.6, make the following changes in the above proof.
In the statement of 3.6, there is no G or p so at the start of the proof, one
must consider all < . The denition of D
i
(u) is the same. We replace
P
G
[T
] with P[T
.
The rest of the proof is almost identical except that instead of referring
to the two conditions on G in the statement of Proposition 3.7, one uses the
fact that P already possesses these qualities by virtue of being
+
-uniform.
35
1
6
2
r
e
v
i
s
i
o
n
:
1
9
9
3
-
1
0
-
1
9
m
o
d
i
f
i
e
d
:
1
9
9
3
-
1
1
-
1
1
References
[1] W. Hodges. Building models by games. LMSST 2. Cambridge University
Press, 1985.
[2] M. Magidor and J.Malitz. Compact extensions of L(Q). Annals of Math.
Logic, 1977.
[3] S. Shelah [Sh 72]. Models with second order properties I: Boolean algebras
with no undenable automorphisms. Annals of Math. Logic 14 (1978),
57-72.
[4] S. Shelah [Sh 73]. Models with second order properties II: On trees with
no undenable branches. Annals of Math. Logic 14 (1978), 73-87.
[5] S. Shelah [Sh 82]. Models with second order properties III: Omitting
types in
+
for L(Q). Archive fur Math. Logik 21 (1981), 1-11.
[6] M. Rubin and S. Shelah [Sh 84]. On the elementary equivalence of au-
tomorphism groups of Boolean algebras, downward Skolem-Lowenheim
theorems and compactness of related quantiers. Journal of Symbolic
Logic 45, 1980, 265-283.
[7] S. Shelah [Sh 89]. Boolean algebras with few endomorphisms. Proc. Am.
Math. 14 (1979), 135-142.
[8] S. Shelah [Sh 107]. Models with second order properties IV: A general
method and eliminating diamond. Annals of Pure and Applied Logic 25
(1983), 183-212.
[9] S. Shelah [Sh 128]. Uncountable constructions. Israel Journal of Mathe-
matics 51 (1985), 273-297.
[10] S. Shelah [Sh 176]. Can you take Solovays inaccessible away? Israel
Journal of Mathematics 48 (1984), 1-47.
[11] S. Shelah [Sh 326]. Vive la dierence! To appear.
36