Anda di halaman 1dari 18

(

9
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
0
5







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
1
-
0
6


NO LIMIT MODEL IN INACCESSIBLE
SH906
SAHARON SHELAH
Dedicated to Michael Makkai
Abstract. Our aim is to improve the negative results i.e., non-existence of
limit models, and the failure of the generic pair property from [5] to inaccessible
as promised there. In [5], the negative results were obtained only for non-
strong limit cardinals.
0. Introduction
Let =
<
> be regular cardinals. A complete rst order theory T may
have (some variant of) (, )-limit model, which, if exists, is unique, see history in
[5] and Denition 0.9. There we prove existence for the theory of linear order and
non-existence for rst order theories which are strongly independent and then just
independent and even the parallel for = 2 (one direction of the so-called generic
pair conjecture). Those non-existence results in [5] were for = 2

, here we deal
with strongly inaccessible . In [6] there are existence results but for measur-
able, and we promise there the non-existence results for strongly inaccessible as
complimentary results.
Let be strongly inaccessible (> [T[) such that
+
= 2

; this for transparency


only.
Here in 1 we prove that for strongly independent T (see Denition 0.2), a strong
version of the generic pair conjecture (see Denition 0.7(2)) holds. We also prove
the non-existence of (, )-limit models, a related property (for all versions of limit
model).
In 2, we also prove this even for independent T. The use of
+
= 2

is just
to have a more transparent formulation of the conjecture. See more on the generic
pair conjecture for dependent T in [4].
We thank Itay Kaplan for much helpful criticism.
{0.4}
Notation 0.1. 1) D

is the club lter on for regular uncountable.


2) S

= < : cf() = .
3) For a limit ordinal let P
ub
() = U : U is an unbounded subset of .
4) T denotes a complete rst order theory.
5) For a model M, ( x, y) L(
M
) and

d
g( y)
M, let (M,

d) = c
g( x)
M :
M [= [ c,

d].
Date: January 6, 2011.
The author would like to thank the Israel Science Foundation for partial support of this research
(Grant No.242/03). The author thanks Alice Leonhardt for the beautiful typing. First Typed -
06/Sept/12. Publ. 906.
1
(
9
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
0
5







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
1
-
0
6


2 SAHARON SHELAH
6) S
n
(A, M) = tp(

b, A, N) : M N and

b
n
N where tp( c, A, N) = ( x, a) :
( x, y) L(
M
), a
g( y)
A and M [= [ c, a].
7) S
n
(M) = S
n
(M, M) and S
<
(M) = S
n
(M) : n .
8)

() = + 2

()
: < and

(
0
).
Recall (as in [5, 2.3])
{0.7}
Denition 0.2. 1) T has the strong independence property (or is strongly inde-
pendent) when : some ( x, y) L(
T
) has it, where:
2) ( x, y) L(
T
) has the strong independence property for T when for every
n < , model M of T and pairwise disjoint nite I
1
, I
2

g( y)
(M) for some
a
g( x)
M we have 1, 2

b I

M [= [ a,

]
if(=2)
.
Remark 0.3. 1) Elsewhere we use (x, y), i.e. the x and y are singletons, but the
proofs are not aected.
2) Also we may restrict ourselves to I
1
, I
2
(M,

d) where L(
T
) such that
(M,

d) is innite, and we may restrict ourselves to I
1
, I
2
such that every

b I
1
I
1
realizing a xed non-algebraic type p S
m
(A, M) with M being ([A[
+
+
0
)-
saturated. The results are not really aected.
{0q.37}
Question 0.4. 1) Assume
2
=
<1
2

1
> [T[, T a complete rst order depen-
dent theory. Is the theory T

1,2
a dependent theory or at least when is T

1,2
a
dependent theory? where
(a) T

1,2
= Th(K
+
1,2
) where
(b) K
+
1,2
= (N, M) : M is a
1
-saturated model of T of cardinality
2
, N a

+
2
-saturated elementary extension of M.
2) Similarly for other properties of T

1,2
; note
1
that this theory is complete if

1
=
2
.
3) When can we prove that T

1,2
does not depend on the cardinals at least for
many pairs?
Remark 0.5. 1) Concerning failure of 0.4(1) see Kaplan-Shelah [3].
2) Any solution of the generic pair conjecture answers positively 0.4(3) for depen-
dent T in the relevant cases.
3) It is known that in 0.4(1) if T extends PA or ZFC then in T

= Th(N, M) we
can interpret the second order theory of
2
.
But may well be that as in Baldwin-Shelah [1]
{0q.44}
Question 0.6. Assume [T[ <
1

2
=
<1
2
, T a complete rst order theory.
For which Ts can we interpret in M K
+
1,2
a model of PA of cardinality
1
by rst order formula or just an L
,
(
T
)-formulas with parameters, the intention
is that we assume
2
is enough larger than
1
which is large enough than [T[; if
2


1
this is trivial.
Recall (from [5, 0.2])
1
let (N

, M

) K
+

1
,
2
for = 1, 2 and let f be an isomorphism from M
1
onto M
2
and let
F = {f : f is a (N
1
, N
2
)-elementary mapping extending f of cardinality
1
}. Now we can
prove that any f F preserve satisfaction for rst order formulas.
(
9
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
0
5







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
1
-
0
6


NO LIMIT MODEL IN INACCESSIBLE SH906 3
{0.14}
Denition 0.7. 0) Let EC

(T) be the class of models M of (the rst order) T of


cardinality . Let EC
,
(T) be the class of -saturated models M EC

(T).
1) Assume > [T[, (we usually assume =
<
) and 2

=
+
, M

EC

(T) is
-increasing continuous for <
+
with M = M

: <
+
EC

+(T), and
M is saturated. The generic pair property (for T, ) says that for some club E of

+
for all pairs < of ordinals from E of conality , (M

, M

) has the same


isomorphism type (we denote this property of T by Pr
2
,
(T)).
2) The generic pair conjecture for >
0
, (usually =
<
) such that 2

=
+
says that for any complete rst order T of cardinality < , T is dependent i it has
the generic pair property for .
3) Let n
,
(T) be min[M

/

=: E has conality [ : E a club of
+
for
and

M = M

: <
+
) as above and = cf() ; clearly the choice of

M is
immaterial.
{0.22}
Remark 0.8. 1) Note that to say n
,
(T) = 1 is a way to say that T has (some
variant of) a (, )-limit model, see 0.9 below. There are other variants of the
Denition of limit.
2) Recall that we conjecture that for =
<
> = cf() > [T[, 2

=
+
we have
n
,
(T) = 1 n
,
(T) < 2

T is dependent. The use of


+
= 2

is just for
clarity. See more in [5], [6], [4].
3) Recall that if = =
<
, then n
,
(T) = 1 means T has a unique saturated
model; (and parallely if > cf() = , strong limit). So we concentrate on the
case n
,
(T) = 1 where < .
{0.26}
Denition 0.9. We dene when M

is a (, )-limit model of T where =


cf() and [T[. In general it means that: letting K

= M : M is a model
of T with universe an ordinal [,
+
), for some function F with domain K and
satisfying M F(M) K we have:
if M

K for <
+
is -increasing continuous and < F(M
+1
)
M
+2
then for some club E of
+
we have:
E cf() = M


= M

.
{0.29}
Remark 0.10. If 2

=
+
we have: n
,
(T) = 1 i T has a (, )-limit model.
(
9
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
0
5







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
1
-
0
6


4 SAHARON SHELAH
1. Strongly independent T
Context 1.1. 1) T is a xed rst order complete theory and C = C
T
a monster for
it; for notational simplicity
T
is relational.
2) We let be a regular uncountable cardinal > [T[; we deal mainly with strongly
inaccessible .
Here for strongly inaccessible and (complete rst order) T with the strong
independence property (of cardinality < ) we prove the non-existence of (, )-
limit models for = cf() < (in Theorem 1.9) and the generic pair conjecture for
and T, in Theorem 1.10 (which shows non-isomorphism). Recall that the generic
pair property speaks on the isomorphism type of pairs of models.
Denition 1.2 gives us a more constructive invariant of (N, M)/

=. Unfortu-
nately it seemed opaque how to manipulate it so we shall use a related but dierent
version, the one from Denition 1.4. Naturally it concentrates on types in one for-
mula ( y, x) witnessing the strong independence property. But mainly gives the
pair (N, M) an invariant P

: < )/D

where P

P(P()). Now always


[P

[ 2
||
and it is easily computable from one P P(), in fact from the invari-
ant inv
4
(M, N) from Denition 1.2, but in our proofs its use is more transparent.
It has monotonicity property and we can increase it.
We need dierent but similar version for the proof of non-existence of (, )-limit
models.
{inv.7}
Denition 1.2. 1) Let E

T
be the following two-place relation on (M, P) : M [= T
and P S
<
(M); let (M
1
, P
1
)E

T
(M
2
, P
2
) i there is an isomorphism h from M
1
onto M
2
mapping P
1
onto P
2
.
2) For models M N we dene (the important case is M N [= T)
(a) inv
1
(M, N) = p S
<
(M) : p is realized in N
(b) inv
2
(M, N) = (M, inv
1
(M, N))/E

T
.
3) If M N are models of T such that the universe of N is , recalling D

is
the club lter on , let:
(a) for any ordinal <
inv
3
(, M, N) = (M , p S
<
(M ) : p is realized by some
sequence from N)/E

T
)
(b) inv
4
(M, N) = inv
3
(, M, N) : < )/D

.
4) If M N are models of T of cardinality then inv
4
(M, N) is inv
4
(f(M), f(N))
for every one-to-one function f from N into (equivalently some f, see 1.3(1),(2)
below)
{inv.8}
Observation 1.3. 0) In Denition 1.2(3) for a club of s below we have M
M and N N and so M N [= T.
1) Concerning Denition 1.2(3), if M N are models of T of cardinality
and f
1
, f
2
are one-to-one functions from N into then inv
4
(f
1
(M), f
1
(N)) =
inv
4
(f
2
(M), f
2
(N)) using the denition 1.2(3)(b).
2) Denitions 1.2(3), 1.2(4) are compatible and in 1.2(4), some f such that f is a
one-to-one function from N to is equivalent to every f such that...
(
9
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
0
5







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
1
-
0
6


NO LIMIT MODEL IN INACCESSIBLE SH906 5
Proof. Straight, e.g. (this argument will be used several times).
1) Let E = < : is a limit ordinal such that M M, N N and
Rang(f

) = Rang(f

) for = 1, 2. So E is a club of and E


f
2
f
1
1
is an isomorphism from f
1
(N) onto f
2
(N), mapping f
1
(M) onto
f
2
(M).
1.3
{inv.10}
Denition 1.4. Assume = ( x, y) L(
T
) and N
1
N
2
are models of T of
cardinality .
1) For one-to-one mapping f from N
2
to and < we dene
inv

5
(, f, N
1
, N
2
) = P P() : there are a


g( x)
N
2
satisfying f( a

)
g( x)

for < such that for every


U the following are equivalent :
(i) U P
(ii) for some

b
g( y)
N
1
we have
< N
2
[= [ a

b]
if(U)
.
2) We let inv

6
(N
1
, N
2
) be inv

5
(, f, N
1
, N
2
) : < )/D

for some (equivalently


every) f as above.
{inv.11}
Claim 1.5. 1) In Denition 1.4(2) we have inv

6
(N
1
, N
2
) is well dened.
2) In Denition 1.4, for , , N
1
, N
2
, ( x, y) as there
(a) the set inv

5
(, f, N
1
, N
2
) has cardinality at most 2
||
(b) if is a one-to-one function from f(N
2
) into mapping f(N
2
) onto
(f(N
2
)) then inv

5
(, f, N
1
, N
2
) = inv

5
(, f, N
1
, N
2
).
Proof. Easy.
1.5
{inv.12}
Denition 1.6. 1) For = ( x, y) L(
T
), a model N of T with universe , a
limit ordinal < and < let
inv

7,
(, N) = P P() : we can nd a
i


g( x)
for < , i < such that
the following conditions on U are equivalent :
(i) U P
(ii) for some

b
g( y)
N we have :
for every i < large enough for every
< we have N [= [ a
i

b]
if(U)
.
2) For = ( y, x) L(
T
) and a model N of T of cardinality let inv

8,
(N) =
inv

7
(, N

) : < )/D

for every, equivalently some model N

isomorphic to N
with universe .
{inv.13}
Observation 1.7. 1) inv

8,
(N) is well dened for N EC

(T) when [T[ + < .


2) In Denition 1.6(1) we have [inv

7,
(, N)[ 2
||+
.
Proof. Easy.
1.7
{inv.21}
Claim 1.8. Assume > [T[ is regular, S is stationary, = ( x, y) and
(a) N
i
: i < ) is a -increasing sequence
(b) N
i
EC

(T)
(
9
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
0
5







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
1
-
0
6


6 SAHARON SHELAH
(c) N = N
i
: i <
(d)

P = P

: < ) where P

P()
(e) f is a one-to-one function from N onto
(f) for every j < for a club of s below there are a
j

N
j+1
f
1
() for
< satisfying
() for every c
g( x)
(N
j
) there is U P

such that < N [=


[ a
j

, c]
if(U)
() for every U P

for some

b
g( y)
(N

) we have < N [=
[ a
j

b]
if(U)
.
Then S : P

inv

7,
(, f(N)) D

+S.
Proof. Straight.
1.8
Now we come to the main two results of this section.
{inv.14}
Theorem 1.9. For some club E of
+
, if
1
,=
2
belong to ES

then M
1
, M
2
are not isomorphic, moreover inv

8,
(M
1
) ,= inv

8,
(M
2
) when :
(a) T has the strong independence property (see Denition 0.2)
(b) =
<
is regular uncountable, > [T[, > = cf() and
+
= 2

(c) M is a saturated model of T of cardinality


+
(d) M

: <
+
) is -increasing continuous sequence with union M,
each of cardinality .
{inv.24}
Theorem 1.10. Assume of 1.9.
1) For some club E of
+
, if
1
<
2
<
3
are from E and

for = 1, 2, 3
then (M
2
, M
1
) (M
3
, M
1
), moreover inv

6
(M
2
, M
1
) ,= inv

6
(M
3
, M
1
) for
some .
2) If M N
0
are models of T of cardinality , then for some elementary extension
N
1
EC

(T) of N
0
we have N
1
N
2
EC

(T) (N
0
, M) ,

= (N
2
, M).
Discussion 1.11. We shall below start with M EC

(T) and a sequence b


i
: i <
) of distinct members such that (

b
i
, y) : i < ) are independent, and like to
nd N, a
i
: i < ) such that M N EC

(T) and the

b
i
: i < ) has a real
aect on the relevant -invariant, in the case of 1.10(1) this is inv

6
(M, N): for a
stationary set of s below it adds something to the -th component in a specic
representation, i.e. assuming f : N is a one-to-one function and we deal with
inv

5
(, f, M, N) : < ); we have freedom about ( a

b
i
) and we can assume

b
g( y)
M

b
i
: i < N [= [ a

b].
But the relevant P

is inuenced not just by say

b
i
: i [, 2
||
)) but also by
later

b
i
s (and earlier

b
i
). To control this we use below a

: < ), S, E such
that we deal with dierent S in an independent way to large extent; this is the
reason for choosing the C

s.
Proof. Proof of 1.9 By the proof of [5, 2] without loss of generality is strongly
inaccessible. Choose Reg
0
, will be needed when we generalize the
proof in 2.
Let U
i
: i < ) be a -increasing sequence of subsets of such that the set
U

i
= U
i
U
j
: j < i has cardinality for each i < and let U

= U
i
: i <
(
9
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
0
5







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
1
-
0
6


NO LIMIT MODEL IN INACCESSIBLE SH906 7
. Let ( x, y) L(
T
) have the strong independence property, see Denition 0.2.
We can choose C

: <
+
) such that C

nacc(), otp(C

) , C

= C

and [ cf() = = sup(C

) and cf() ,= otp(C

) < .
[See [7] but for completeness we show this; by induction on <
+
we choose
C

: < ) such that:


(a) the relevant demand holds
(b) if = + 1, C , (i C)(C

i
= C i) and otp(C) < then for some
i (, ) we have C

i+1
= C.
As =
<
this is easy but we elaborate. For = 0 trivial for limit obvious.
Assume = + 1 let

= ,

= and C

i
: i <

) has been dened.


First, we choose C

i
. If cf(

) ,= let C

= , so assume cf(

) = then
necessarily cf() = .
Let

: < ) be increasing with limit , and choose

+ ) by
induction on < such that C

: < .
Lastly, let C

:=

: < . So C

has been dened in any case.


Now let C

= C : otp(C) < and C C

= C , so [C[ ,
also C, so let C

+i
: i (0, )) list C

possibly with repetitions. So we have


dened C

: < ), so have carried the induction.]


Let S

= : =
+
for some < . Let E

, C

: < ) be such that:

1
(a) C

(b) C

= C


(c) otp(C

)
(d) E

is the club < : < =

of
(e) otp(C

) = i E

(f) if S := E

then = sup(C

)
(g) if E

and i < then [ U

i
[ = [[.
[Why can we choose? By induction on the cardinal
0
we choose C

: <

)
and E

= E

such that the relevant demands hold and: if = 2


1
and C is a
subset of S

1
of order type < satisfying C C

= C then for some


S

(
1
,

) we have C

= C. Why this extra induction hypothesis help? As


arriving to S so =

let
i
: i < ) be an increasing sequence of cardinals
with limit

= and we choose
i
(
i
,
2

i ) S

by induction on i < such


that C
,i
=
j
: j < i and the let C

=
i
: i < .]
We shall prove that

2
if
2
below holds, then there is a such that
2
holds where:

2
(a) <
+
, i <
(b) f is a one-to-one function from M

into U

i
= U
j
: j < i
(c) E E

is a club of such that E f(M

) f(M

)
(d)

P = P

: S)
(e) P

P() and P

and P

2
P
,

where
2
2
note that P
,1
,i
, P
,2
,i
are the families of sets we like to ignore as they are inuenced by our
choices for
1
S\{}, so we work to have them families of bounded subsets of .
(
9
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
0
5







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
1
-
0
6


8 SAHARON SHELAH
() P
,0

= A : sup(A) = and A [, 2

) : C

,
() P
,1

= P
,0
1
:
1
S ,
() P
,2

= A : for some
1
( + 1) we have
A [, 2

) : C
1

(f) if
1
<
2
are from S then
() A P
1
A P
2
() A P
2
A
1
P
1
,
() for any S the family P
,1

P
,2

is a set of bounded
subsets of ; (this follows)
(g)

b
,U
M

for E S, U P

are such that

b
1,U1
=

b
2,U2
U
1
P
1
U
2
P
2

1
=
2
U
1
= U
2

2
() (,
+
)
() there are a

for < such that for a club of E, if


S then the following conditions on U are equivalent:
(i) U P

(ii) for some



b
g( y)
M

we have: for every < ,


M

[= [ a

b] i U
(iii) clause (ii) holds for

b =

b
,U
and U P
,i
[Why? For each ES let U
,
: < [P
,i
[ 2
||
) list P

and let

b
,
:=

b
,U
,
.
Let
= ( x

b
,
)
if(U
,
)
: < , E and < [P
,i
[
( x

b) : < ,

b
g( y)
(M

) and for no
E, < [P

[ do we have

b =

b
,
.
As ( x, y) has the strong independence property, recalling that by clause (g) of
2
the sequence

b
,
: E S and < [P

[) is with no repetitions, clearly is


nitely satisable in M

, but M is
+
-saturated, M

M and [[ = hence we
can nd a


g( x)
M for < such that the assignment x

( < ) satises
in M. Lastly, choose (,
+
) such that a

: < M

.
Now check recalling P

for S.]
Note

3
in
2
if h is a one-to-one mapping from M

into U extending f then for


some club E of if for every S E we have ( < )( < h( a

)
g( x)
) and so for every U the conditions (i), (ii), (iii) from
2
are
equivalent.
Next we can choose

f such that

3
(a)

f = f

: <
+
)
(b) f

is a one-to-one function from M

into U
otp(C

)
(c) if C

then f

.
Now
(
9
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
0
5







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
1
-
0
6


NO LIMIT MODEL IN INACCESSIBLE SH906 9

4
for every <
+
there is

P

= P

: S) such that
(i) P

P() are as in
2
(e), (f) above
(ii) for every , for a club of s fromS we have P

/ inv

7,
(, f

(M

)).
[Why? For every and (, ) we have inv

7,
(, f

(M

)) is a subset of
P(P()) of cardinality 2
||
. As the number of s is , by diagonalization we
can do this: let + 1 =

<
u

and u

[ +1]
<
increasing continuous for < ;
moreover, < u

= and u

and [u

[ [[. By induction on
(, ) S choose P

<3
P
,

which includes P

: u

SP
,2

and
satises P
,0

P(P
,0
,i
))P P
,0

: P inv

7,
(, f

(M

)), u

.
Note that for each the set < : u

contains an end-segment of
hence a club of as required.]
Now choose pairwise distinct

b
,U

g( y)
(M
0
) for E

, U P
,0

5
for every

<
+
for some (,
+
) and a


g( x)
M

for <
the condition in clause () of
2
holds with

P

here standing for



P there
and the

b
,U
chosen above.
[Why? By
2
.]

6
let E = <
+
: is a limit ordinal such that for every

< there
is < as in
5
.
Clearly E is a club of
+
.

7
if
1
<
2
are from E S

then M
1
, M
2
are not isomorphic.
[Why? Let

= min(C

1
). We consider

P

which is from
4
. On the one
hand < : P

/ inv

7,
(, f
1
(M
1
)) contains a club by
4
(ii). Note that
f

: C

1
) is -increasing sequence of functions with union f
2
.
On the other hand choose an increasing
i
: i < ) with limit
2
satisfying
0
=
0,
1
=
1
such that (

,
1+i
,
1+i+1
) are like (

, , ) in
5
for each i < and
i (1, )
i
C

2
. Now by 1.8, < : P

inv

7,
(, f
2
(M
2
)) contains a
club. Hence by the last sentence and the end of the previous paragraph M
1
M
2
as required.]
So we are done.
1.9
Remark 1.12. We can avoid using C

and also C

(e.g. using A P
,0


otp(A) = ) but seems less transparent.
Proof. Proof of 1.10 Similar but easier (for regular not strong limit (but 2

>
2
<
) also easy), or see the proof of 2.9.
1.10
(
9
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
0
5







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
1
-
0
6


10 SAHARON SHELAH
2. Independent T
We would like to do something similar to 1, but our control on the relevant
family of subsets of is less tight. We control it to some extent by using the
completion of a free Boolean algebra.
{2d.1}
Context 2.1. T a complete rst order theory, (x, y) has the independence property
(of course the existence of such follows from the strong independence property
but is weaker).
We continue [5, 2.1-2.12], but we do not rely on it.
{2d.4}
Denition 2.2. For a set I let
(a) B = B
I
be the Boolean Algebra generated by e
t
: t I) freely,
(b) B
c
I
is the completion of B
(c) for J I let B
c
I,J
be the complete subalgebra of B
c
I
generated by e
s
: s
J
(d) let uf(B
c
I
) be the set of ultralters on I.
{2d.7}
Claim 2.3. Assume
(a) M [= T
(b)

b
t

g( y)
M for t I
(c) (x,

b
t
) : t I) is an independent sequence of formulas.
Then there is a function F from
g( y)
M to B = B
c
I
such that
() F(

b
t
) = e
t
() for every ultralter D of B there is p = p
D
= p
F,D
S

(M), in fact, a
unique one, such that for every

b
g( y)
M we have (x,

b) p F(

b) D.
Remark 2.4. 1) Note that the mapping D p
D
is not necessarily one to one, but
D
1
e
t
: t I ,= D
2
e
t
: t I p
D1
,= p
D2
.
2) If I = I
1
I
2
, I
1
I
2
= and [I
2
[ = [I
1
[
0
then we can nd a mapping F from
g( y)
M onto (not just into) B = B
c
I1
such that clause (), () are satised.
Proof. Clearly P(M) is a Boolean algebra and (M,

b
t
) : t M generates freely
a subalgebra of P(M) which we call B

. So there is a homomorphism h from B

into B mapping (M,

b
t
) to e
t
(moreover h is unique and is an isomorphism from
B

onto B
I
B
c
I
). So h is a homomorphism from B

P(M) into B
c
I
, which is
a complete Boolean algebra hence there is a homomorphism h
+
from the Boolean
algebra P(M) into B
c
extending h.
Lastly, dene F :
g( y)
M B
c
by F(

b) = h
+
((M,

b)). Now check.


2.3
{2d.14}
Conclusion 2.5. Assume from 2.3 and
(a) I = is regular uncountable
(b) [M[ U
(c) D

is an ultralter of B
c
I
for <
(d) U [M[ is unbounded in .
Then we can nd a

: < ) and N such that


(
9
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
0
5







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
1
-
0
6


NO LIMIT MODEL IN INACCESSIBLE SH906 11
() M N
() [N[ U
() a

N for <
() a

realizes p
D
S

(M).
Remark 2.6. Conclusion 2.5 is easy but intended to clarify how we shall use the
ultralters, so is quoted toward the end of the section.
Proof. Should be clear.
2.5
{2d.21}
Discussion 2.7. Note that compared to 1 instead x, y, a

we have x, y, a

.
Compared to 1, we have less control over tp(a, M, N) : a N. There, for the
sequences

b of M which are not among

: < , we can demand N [= [ a

b]
for < so tp

( a

, M, N) can be clearly read. Here the complete Boolean Algebra


B
c
I
is helping, a small price is that we need >
0
.
In order to try to keep track of what is going on we shall use only tp(a

, M, N)
of the form p
D
for ultralter D on B
c
I
. Further, we better have, e.g. a nice function
from

2 to uf(B
c
I
) such that (e

()) () = 1.
A possible approach is: we dene M
,u
: T des(), u P(n

)) as in
[8, 3] and we dene D

uf(B
c
M) such that M

[e
()

] and

uf(B
c
).
We need some continuity so each e D

(e B
c
) depends on u
e
for some
small u
e
.
{3e.7}
Theorem 2.8. In Theorem 1.9 it suces to assume

which means clauses


(b),(c),(d) of and
(a)

T has the independence property.


{3e.14}
Theorem 2.9. In Theorem 1.10 it suces to assume

of 2.8.
Proof. Proof of 2.8 Just combine the proofs of 1.9 from 1 and 2.9 below.
2.8
Proof. Proof of 2.9 As in the proof of 1.9 we can assume is strongly inaccessible
though the proof is just easier otherwise. We let

1
(a) E

= < : =

, a club of
(b) S

=
+
: <
(c) choose a regular uncountable <
and let

2
(a) S = E

: cf() = = S

and
(b) let

C be as in
1
of the proof of 1.9, in particular

C =
C

: < ), C

, otp(C

) , C

= C


and S = sup(C

) otp(C

) =
and S sup(C

) =
(c) for S let A

= [, 2

] : C

.
Let D

be an ultralter of B
c

such that e

/ D

for < .
Now for

2 we choose D

such that
(
9
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
0
5







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
1
-
0
6


12 SAHARON SHELAH

3
(a) D

is an ultralter of B
c

(b) if e D

B
c

belongs to B
c
,
1
{0}
(see 2.2, the completion of
the subalgebra of B
c

generated by e

: () = 0) then e D

.
(c) if < and () = 1 then e

.
So

4
(a) if

2 is constantly zero then D

= D

(b) e

() = 1 for < ,

2.
Now let =

: < ) be a sequence of members of



2 and below we shall be
interested mainly in the case = S.
Dene

5
for e B
c

and we let Y

,e
:= < : e D

6
P
,
:= Y

,e
: e B
c

.
Now what can we say on P
,
for S ? As we can consider e e

: [, 2

),
clearly

< :

() = 1 : [, 2

P
,
P().
This may be looked at as a lower bound of P
,
. Naturally we try to get also an
upper bound to P
,
; now note

8
if e B
c

then Y

,e
= Y

,e
.
Now dene (recalling A

is from
2
(c))

9
is the set of of the form

: < ) such that:


(a)



2 for every < ,
(b) if

() = 1 then ( S)[ < 2

],
(c) if S and u [, 2

) is countable then A

: if u then

() = 0 is of cardinality .
Also (by our knowledge of the completion of a free Boolean algebra, B
c

satises
the c.c.c.) for every e B
c

we can choose u
e
such that:

1
(a) u
e
is countable
(b) e B
c
,ue
.
So by clause (b) of
3
clearly

2
if , e B
c

, < S and u
e

1

0 then e D

e D

hence

3
if , e B
c

and S then Y

,e
< : u
e

1

0.
Next

4
for
(a) let D
,
be the lter on generated by A

< : u
1

0
and > : < and u is countable
(b) let I
,
be the dual ideal.
(
9
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
0
5







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
1
-
0
6


NO LIMIT MODEL IN INACCESSIBLE SH906 13
Clearly

5
(a) if , S and [, 2

) then < :

() ,= 0 is
a bounded subset of ,
(b) if and S then D
,
is a uniform
1
-complete lter on
(recalling cf() = >
0
as S) and / D
,
.
[Why? See
9
.]
Now by
3
we have e B
c

,e
D
,
so recalling
8
we have
e B
c

Y

,e
= mod D
,
hence

6
P
,
X : X D
,
or X D
,
.
Now

1
if S then we can nd

A

for < 2
2

such that:
(a)

A

= A

: [, 2

))
(b) A

is an unbounded subset of A

(c) D

, I

are well dened, i.e. D

when we let
() D

be the
1
-complete lter of subsets of generated by
A

: [, 2

) and <
so A

for <
() I

= B : B D

, i.e. the dual ideal


(d) moreover if
1
,=
2
are < 2
2

, then
A

: [, 2

) D

.
[Why
1
holds? As [A

[ = [[ is a strong limit cardinal of conality >


0
clearly
= [A

[ = [A

[
0
hence by [2] there is a sequence B

: [, 2

)) of subsets
of A

such that any non-trivial Boolean combination of countably many of them


has cardinality . Let U

: < 2
2

) be a sequence of pairwise distinct subsets of


[, 2

) each of cardinality 2
||
no one included in another and let A

: [, 2
||
))
list B

: U

.
Now check.]

2
in
1
it follows that
(e) for every P P() for at most one < 2
2

we have
A

: [, 2

) P D

3
for every

= () : S) 2
2

: S there is =

such that:
(a)

so

=
,
: < )
(b) if S, [, 2

) then :
,
() = 1 = A

A
()

.
[Why? Just read the denition of in
9
and

A

in
1
.]

4
if S then D

,
I

,
= D
()

I
()

.
[Why? Easy, recalling
5
(a).]
(
9
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
0
5







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
1
-
0
6


14 SAHARON SHELAH

5
if () <
+
and

P

= P

: S), for < () where P

P(P())
has cardinality 2

for S, < () then we can nd



= () :
S) 2
2

: S such that for every < () the following set


is not stationary: S
,
= S: for some P P

we have A
()

:
[, 2

) P D
,
I
,
.
[Why? Let u

: < ) be an increasing continuous sequence of subsets of ()


with union () such that [u

[ [[ for < . Now for each S, the family


P

: u

is a family of [u

[ 2

subsets of P().
Now by clause (e) of
1
for each S, u

, P P

let
,,P
< 2
2

be such
that: if for some < 2
2

we have A

: [, 2

) P D

then
,,P
is the rst such . Choose () < 2
2

which does not belong to


,,P
: u

and P P

.
So let =
():S
be as in
3
now is as required by
2
,
3
,
4
.
Let us elaborate, why is it as required in
5
?
First, clearly



2 for < . Second, x < (), then there is <
such that u

, so it suces to show that, for any S, we have /


S
,
. So assume P P

satises clause (e) of


1
, and we should prove that
[A
()

: [, 2

) P D
,
I
,
]; but if for some < 2
2

we have
A

: [, 2

) P D
,
I
,
then necessarily =
,,P
,= (),
contradiction to
1
.]

6
if M

: ()) is a -increasing continuous and M

EC

(T) and


g( y)
(M
0
) for < are such that (x,

) : < ) is independent,
then we can nd N such that
() M
()
N EC

(T)
() if N N

EC

(T) and < () then


3
inv

6
(M

, N

) / inv

6
(M
1
, M
2
) :
() and
1
<
2
().
[Why? Without loss of generality the universe of M
()
is U
1
[]

such that U
1
has cardinality . Let u

: < ) be as in the proof of


5
.
For (1) < (2) () let P
(1),(2)

= inv
5

(, id
N
(2)
, M
(1)
, M
(2)
), see De-
nition 1.4, clearly inv

6
(M
(1)
, M
(2)
) = P
(1),(2)

: < )/D

. So it is enough
4
to
nd N and sequence a

: < ) of elements of N such that M


()
N, [N[ =
and for each (0) (), for every S except non-stationarily many, the family
< : N [= [a

b] :

b
g( y)
(M
(0)
)
is not in P

:= P
(1),(2)

: (1) < (2) () are from u

.
We choose

= () : S) as in
5
; let =

, see
3
, so recalling
3
clearly
D

: < ) is well dened. Now for each < letting F be from 2.3 for the model
M
()
and the sequence (x,

) : < ), let p

(M
()
) be such that for every

b
g( y)
(M
()
) we have (x,

b) p

F(

b) D

so (x,

b) p

F(

b) / D

.
So by 2.5 we can nd an elementary extension N of M
()
and a

N for <
such that a

realizes p

, and without loss of generality N has universe such that


3
really any pregiven set of forbidden inv

6
is O.K. and can make it work for inv

6
(N, N

)
for every < ().
4
can demand <
>
(( + 1)) if (N\M
()
) [, +) is innite for every < .
(
9
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
0
5







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
1
-
0
6


NO LIMIT MODEL IN INACCESSIBLE SH906 15
[N[ has cardinality . Concerning inv
6

our demand concerns what occurs for a


club of < for this. Let E E

be a club of such that < E a

N.
Now in
6
() we promise something (given N) on every N

such that ..., so let


N N

EC

(T), and without loss of generality the universe of N

is and
let S E. For any () by
5
,i.e. by the choice of

,

above there is a
club E

E of such that for any S E

, the set S
,
from
5
is disjoint
to E

, hence the set P


,
:= < : N

[= [a

b] :

b
g( y)
(M

) does not
belong to P
(1),(n)

: (1) < (2) () are from u

so we are done.]

7
if M

: <
+
) is as in

then for some club E of


+
, we have: if

1
<
2
,
1
<
2
are from E and
2
,=
2
then
(M
2
, M
1
) (M
2
, M
1
).
[Why? For every <
+
we apply
6
to M

: ) and get N

as there so
M

EC

(T). As M = M

: <
+
is saturated, without loss of
generality N

M hence for some

<
+
we have N

.
Let E = <
+
: a limit ordinal such that <

< .
Let
1
<
2
,
1
<
2
be from E such that
2
,=
2
and we shall prove
that (M
2
, M
1
) is not isomorphic to (M
2
, M
1
). By symmetry without loss
of generality
2
<
2
and let () = max
2
,
1
so () <
2
. Now we ap-
ply
6
with M

: ()), N, N

, ,
1
,
2
,
2
here standing for M

:
()), N, N

, (0), (1), (2) there so we are clearly done.


2.9
(
9
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
0
5







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
1
-
0
6


16 SAHARON SHELAH
3. Private Appendix
Denition 3.1. 1) We say that T has the PO-property when some ( x, y) has it
where:
2) We say ( x, y) has the PO-property in T when: for every partial order I and for
some model M of T and a
t

g( y)
M for t M we have:
(a) if s, t I has no common upper bound then M [= ( x)(( x, a
s
)( x, a
t
))
(b) if n < and t
0
, . . . , t
n1
I has a common upper bound then ( x)

<n
(x, a
t

).
{3n.7}
Claim 3.2. In Th(M, N) we can interpret by L
|T|
+
,
(
T
)-formulas with param-
eters, a Boolean algebra B isomorphic to B

and can interpret qualication over


ideals of B when :
(a) M N are models of T
(b) N is |M|
+
-saturated
(c) B

is a complete Boolean Algebra of cardinality


(d) (x, a

) : < ) is an independent sequence of formulas in M


(e) =
0
(for simplicity).
Remark 3.3. Does this mean to imply interpretations of large models of PA?
Proof. Let B = B
c

and let d

: < ) list the elements of B, see 2.2. Let B


M
be the Boolean algebra of subsets of M denable by formulas with parameters,
i.e. (M,

b) : (x, y) L(
T
) and

b
g( y)
M and let B

be the subalgebra
generated by (M, a

) : < . Let h be the unique homomorphism from B

M
into B
c

mapping (M, a

) to d

; exists as (M, a

) : < ) is an independent
family of subsets of M. As B
c

is a complete Boolean Algebra we can nd an


extension of h to a homomorphism from B
M
into (hence onto) B
c

.
Moved from end of 1,pg.9:
Discussion: We may prefer to have an invariant which speaks only on one model.
Denition 3.4. Let ( be regular [T[, (x, y) as in xxx).
1) If N EC
T
(), f : [N[ and < and P P() we say (N, f, , ) accept
P when: there is a witness a
,
: T , < ) such that
(a) T T

where T T

i T
>
is non-empty, closed under initial
segments, with no innite branch
(b) for every

b
g( y)
N for some U P for almost every T we have
M [= [a
,
,

b]
if(U)
.
FILL
Moved from Denition 1.2,pg.3: 3) If N is a model of T with universe let
inv
3
(M) = inv
2
(M, N) : M N has universe an ordinal < )
inv
4
(M) = inv
3
(M)/D

recalling D

the club lter.


(
9
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
0
5







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
1
-
0
6


NO LIMIT MODEL IN INACCESSIBLE SH906 17
4) If N is a model of cardinality let inv
4
(N) be inv
4
(N

) for any equivalent some


model N

with universe isomorphic to .


We rst deal with the explicit case (and use simplifying) cardinal arithmetic
assumptions
5
. Moved from pgs.10,11:
Remark 3.5. Assume
(a) is strongly inaccessible of conality
(b) S < : =
0
is strong limit is stationary.
Then we can nd

D such that
() D = D
B
: B )
() D
B
is an ultralter of B
c

()
1
if B

for < , = 1, 2 and B


1
1
,= B
2
2
for
1
,
2
< then for a club of
S we have:
< : Y : for some e B

we have ( < )( Y e
D
1
B
) , = y : for some e B
c
we have ( < )( y e D
2
B

()
2
if P
,
P() for < 2

then for some



D = D

: < ), for every


S we have < : e D

: e B
c

/ P
,
: < 2

.
Proof. Choose D

such that

1
D

is an ultralter on B
c

disjoint to e

: < .
Let be such that

2
[?? ] (a) =
1,2
:
1
<
2
are from S)
(b)
1,2
is a mapping from [
2
, 2
2
) to [
1
, 2
1
)
(c) the s commute
(d) if (S ) has not last inverse limit.
For each S we choose

such that

2
?? (a)

=
,
: [, 2

))
(b)
,


2
(c) if
1
<
2
are form S and
1,2
(
2
) =
1
then
1,1

1
:

2,2

2
let , S

, E, S, c be as in the proof of 1.9

3
if =

: < ),



2 for < then let D
,
be the
1
-complete lter
on generated by < :
1

0 u : u is countable.
Now we can choose

such that

5
(a)

A
: A )
(b)

A


2
(c) if =
0
< and A
n
, A
n
: n < ) are pairwise distinct
then
1
A0
1
1
1+n
1 : n < .
Hence
5
if 2

>
+
use the game formulation
(
9
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
1
-
0
5







m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
1
-
0
6


18 SAHARON SHELAH
if A

for < and B P()A

: < then
1
B
1 , = , mod
D
<A
:<>
.
So D
B
: B ) are as required in ().
Proof. Proof of ()
2
The proof is similar. FILL.
We choose

by induction on such that ...


References
[1] John T. Baldwin and Saharon Shelah. Second-order quantiers and the complexity of theories.
Notre Dame Journal of Formal Logic, 26:229303, 1985. Proceedings of the 1980/1 Jerusalem
Model Theory year.
[2] Ryszard Engelking and Monika Kar lowicz. Some theorems of set theory and their topological
consequences. Fundamenta Math., 57:275285, 1965.
[3] Itay Kaplan and Saharon Shelah. A counterexample to existence of indiscernibles and other
examples. preprint.
[4] Saharon Shelah. A dependent dream and recounting types.
[5] Saharon Shelah. Dependent T and Existence of limit models. Tbilisi Mathematical Journal,
submitted. math.LO/0609636.
[6] Saharon Shelah. Dependent theories and the generic pair conjecture. Communications in Con-
temporary Mathematics, submitted. math.LO/0702292.
[7] Saharon Shelah. Remarks on squares. In Around classication theory of models, volume 1182
of Lecture Notes in Mathematics, pages 276279. Springer, Berlin, 1986.
[8] Saharon Shelah. Antihomogeneous Partitions of a Topological Space. Scientiae Mathematicae
Japonicae, 59, No. 2; (special issue:e9, 449501):203255, 2004. math.LO/9906025.
Einstein Institute of Mathematics, Edmond J. Safra Campus, Givat Ram, The He-
brew University of Jerusalem, Jerusalem, 91904, Israel, and, Department of Mathe-
matics, Hill Center - Busch Campus, Rutgers, The State University of New Jersey, 110
Frelinghuysen Road, Piscataway, NJ 08854-8019 USA
E-mail address: shelah@math.huji.ac.il
URL: http://shelah.logic.at

Anda mungkin juga menyukai