Sei sulla pagina 1di 20

(

8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


WHEN FIRST ORDER T HAS LIMIT MODELS
SH868
SAHARON SHELAH
Abstract. We to a large extent sort out when does a (rst order complete
theory) T have a superlimit model in a cardinal . Also we deal with related
notions of being limit.
Date: May 26, 2010.
I would like to thank Alice Leonhardt for the beautiful typing. First Typed - 04/June/23.
1
(
8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


2 SAHARON SHELAH
Anotated Content
0 Introduction
[We give backgroup and the basic denitions. We then present existence
results for stable T which have saturated or closed saturated models.]
1 On countable superstable not
0
-stable
[Consistently 2
1

2
and some such (complete rst order) T has a super-
limit (non-saturated) model of cardinality
1
.]
2 A strictly stable consistent example
[Consistently
1
< 2
0
and some countable stable not superstable T, has
a (non-saturated) model of cardinality
1
which satises some relatives of
being superlimit.]
3 On the non-existence of limit models
[The proofs here are in ZFC. If T is unstable it has no superlimit models
of cardinality when
1
+ [T[. For unsuperstable T we have similar
results but with few exceptional cardinals on which we do not know:
<
0
which are <

. Lastly, if T is superstable and [T[ + 2


|T|
then T has a superlimit model of cardinality i [D(T)[ i T has
a saturated model. Lastly, we get weaker results on weaker relatives of
superlimit.]
(
8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


WHEN FIRST ORDER T HAS LIMIT MODELS SH868 3
0. Introduction
(0A) Backgroup and Content
For a rst order complete T, we know : T has a saturated model of T of
cardinality , that is, it is :
<
[D(T)[ or T is stable in . What if
we replace saturated by superlimit (or the relatives limit,weakly, strongly limit)?
Those notions were suggested for a.e.c. in [Sh:88, 3.3] or see in the revised version
[Sh:88r, 3.6] and see [Sh:h] or here in 0.6, but what do they mean when the class is
elementary? Let EC

(T) be the class of models M of T of cardinality .


If there is a saturated M EC

(T) we have considerable knoweldge as men-


tioned in [Sh:88r] by [Sh:c], see 0.5. E.g. for superstable T in 2
|T|
there is
a superlimit model (the saturated one). It seems a natural question on [Sh:88,
3.3] whether it exhausts the possibilities of (, )-superlimit and (, )-superlimit
models for elementary classes.
We rst deal with the consistency results from [Sh:100].
Clearly the cases of the existence of such models of a (rst order complete)
theory T where there are no saturated (or special) models are rare, because even
the weakest version of Denition [Sh:88, 3.3] = [Sh:88r, 3.6] or here 0.6 for implies
that T has a universal model of cardinality , which is rare (see Kojman Shelah
[KjSh:409] which includes earlier history and recently Djamonza [Mirar]).
So the main question seems whether there are such cases at all. We naturally
look at some of the previous cases of consistency of the existence of a universal
model (for <
<
), i.e., those for =
1
.
In 1 we deal with the case of the countable superstable T
0
from [Sh:100] which is
not
0
-stable. By [Sh:100] consistently
1
< 2
0
and for some T

0
T
0
of cardinality

1
, PC(T

0
, T
0
) is categorical in
1
. We use this to get the consistency of T
0
has a
superlimit model of cardinality
1
and
1
< 2
0
.
In 2 for some stable not superstable countable T
1
we have a parallel but weaker
result. We relook at the old consistency results of some PC(T

1
, T
1
), [T

1
[ =
1
>
[T
1
[, is categorical in
1
from [Sh:100]. From this we deduce that in this universe,
T
1
has a strongly (
1
,
0
)-limit model.
It is a reasonable thought that we can similarly have a consistency result on the
theory of linear order, but this is still not clear.
In 3 we show that if T has a superlimit model in [T[ +
1
then T is stable
and T is superstable except possibly under some severe restrictions on the cardinal
(i.e., <

and <
0
). We then prove some restrictions on the existence of
some (weaker) relatives.
Summing up our results on the main notion, superlimit, by 1.1 + 3.1:
{n0.1}
Conclusion 0.1. Assume [T[+

. Then T has a superlimit model of cardinality


i T is superstable and [D(T)[.
In subsequent work we shall show that for some unstable T (e.g. the theory of
linear orders) if =
<
> = cf(), then T has a medium (, )-limit model,
whereas if T has the independence property even weak (, )-limit models do not
exist; see [Sh:877] and more in [Sh:900], [Sh:906], [Sh:950], [Sh:F1054].
We thank Alex Usvyatsov for urging us to resolve the question of the superlimit
case and John Baldwin for comments and complaints.
(0B) Basic Denitions
(
8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


4 SAHARON SHELAH
{y.1}
Notation 0.2. 1) Let T denote a complete rst order theory which has innite
models but T
1
, T

, etc. are not necessarily complete.


2) Let
T
= (T),
M
= (M) be the vocabulary of T, M respectively.
{y.2}
Denition 0.3. 1) For any T let EC(T) = M : M is a
T
-model of T.
2) EC

(T) = M EC(T) : M is of cardinality .


3) For T T

let
PC(T

, T) = M
T
: M is model of T

PC

(T

, T) = M PC(T

, T) : M is of cardinality .
4) We say M is -universal for T
1
when it is a model of T
1
and every N EC

(T)
can be elementarily embedded into M; if T
1
= Th(M) we may omit it.
5) We say M EC(T) is universal when it is -universal for = |M|.
We are here mainly interested in
{y.3}
Denition 0.4. Given T and M EC

(T) we say that M is a superlimit or -


superlimit model when : M is universal and if <
+
is a limit ordinal, M

: )
is -increasing continuous, and M

is isomorphic to M for every < then M

is isomorphic to M.
{y.4}
Remark 0.5. Concerning the following denition we shall use strongly limit in
2.14(1), medium limit in 2.14(2).
{y.5}
Denition 0.6. Let be a cardinal [T[. For parts 3) - 7) but not 8), for
simplifying the presentation we assume the axiom of global choice and F is a class
function; alternatively restrict yourself to models with universe an ordinal [,
+
).
1) For non-empty :
0
< and is regular and M EC

(T) we say
that M is a (, )-superlimit when : M is universal and
if M
i
: i ) is -increasing, M
i

= M for i < and
then M
i
: i <

= M.
2) If is a singleton, say = , we may say that M is (, )-superlimit.
3) Let S
+
be stationary. A model M EC

(T) is called S-strongly limit or


(, S)-strongly limit when for some function: F : EC

(T) EC

(T) we have:
(a) for N EC

(T) we have N F(N)


(b) if S is a limit ordinal and M
i
: i < ) is a -increasing continuous
sequence
1
in EC

(T) and i < F(M


i+1
) M
i+2
, then M

= M
i
:
i < .
4) Let S
+
be stationary. M EC

(T) is called S-limit or (, S)-limit if for


some function F : EC

(T) EC

(T) we have:
(a) for every N EC

(T) we have N F(N)


1
no loss if we add M
i+1

= M, so this simplies the demand on F, i.e., only F(M) is required
(
8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


WHEN FIRST ORDER T HAS LIMIT MODELS SH868 5
(b) if M
i
: i <
+
) is a -increasing continuous sequence of members of
EC

(T) such that F(M


i+1
) M
i+2
for i <
+
then for some closed un-
bounded
2
subset C of
+
,
[ S C M


= M].
5) We dene
3
S-weakly limit, S-medium limit like S-limit, S-strongly limit
respectively by demanding that the domain of F is the family of -increasing con-
tinuous sequence of members of EC

(T) of length <


+
and replacing F(M
i+1
)
M
i+2
by M
i+1
F(M
j
: j i + 1)) M
i+2
.
6) If S =
+
then we may omit S (in parts (3), (4), (5)).
7) For non-empty : and is regular, M is (, )-strongly limit
4
if
M is <
+
: cf() -strongly limit. Similarly for the other notions. If we do
not write we mean = |M|.
8) We say that M K

is invariantly strong limit when in part (3), F is just a


subset of (M, N)/

=: M N are from EC

(T) and in clause (b) of part (3) we


replace F(M
i+1
) M
i+2
by (N)(M
i+1
N M
i+2
((M, N)/

=) F).
But abusing notation we still write N = F(M) instead ((M, N)/

=) F. Similarly
with the other notions, i.e., we use the isomorphism type of

MN).
{y.5c}
Observation 0.7. 1) Assume F
1
, F
2
are as above and F
1
(N) F
2
(N) (or F
1
(

N)
F
2
(

N)) whenever dened. If F
1
is a witness then so is F
2
.
2) All versions of limit models implies being a universal model in EC

(T).
3) Obvious implication diagram: For non-empty : is regular and
{y.6}
stationary S
1
<
+
: cf() :
superlimit = (, : regular)-superlimit

(, )-superlimit

S
1
-strongly limit

S
1
-medium limit, S
1
-limit

S
1
-weakly limit.
{y.7}
Lemma 0.8. Let T be a rst order complete theory.
1) If is regular, M a saturated model of T of cardinality , then M is (, )-
superlimit.
2) If T is stable, and M is a saturated model of T of cardinality
1
and
= : (T) and is regular), then M is (, )-superlimit (on
2
alternatively, we can use as a parameter a lter on
+
extending the co-bounded lter
3
Note that M is (, S)-strongly limit i M is ({, cf() : S})-strongly limit.
4
in [Sh:88r] we consider: we replace limit by limit

if F(M
i+1
) M
i+2
, M
i+1

F(M
j
: j i + 1) M
i+2
are replaced by F(M
i
) M
i+1
, M
i
F(M
j
: j i) M
i+1

respectively. But (EC(T), ) has amalgamation.


(
8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


6 SAHARON SHELAH
(T)-see [Sh:c, III,3]). (Note that by [Sh:c] if is singular or just <
<
and
T has a saturated model of cardinality then T is stable (even stable in ) and
cf() (T)).
3) If T is stable in and = cf() then T has an invariantly strongly (, )-
limit model.
Proof. 1) Let M
i
be a -saturated model of T for i < and M
i
: i < ) is -
increasing and M

i<
M
i
. Now for every A M

:= of cardinality < there is


i < such that A M
i
hence every p S(A, M

) is realized in M
i
hence in M

;
so clearly M

is -saturated. Remembering the uniqueness of a -saturated model


of T of cardinality we nish.
2) Use [Sh:c, III,3.11]: if M
i
is a -saturated model of T, M
i
: i < ) increasing
cf() (T) then

i<
M
i
is -saturated.
3) Let K
,
=

M :

M = M
i
: i ) is -increasing continuous, M
i
EC

(T)
and (M
i+2
, c)
cMi+1
is saturated for every i < . Clearly

M,

N K
,
M


=
N

. Also for every M EC

(T) there is N such that M N and (N, c)


cM
is
saturated, as also Th((M, c)
cM
) is stable in ; so there is F : EC

(T) EC

(T)
such that M F(M) and (F(M), c)
cM
is saturated; such F witness the desired
conclusion.
0.8
{y15}
Denition 0.9. 1) For a regular uncountable cardinal let

I[] = S : some
pair (E, a) witnesses S

I[], see below.
2) We say that (E, u) is a witness for S

I[] i:
(a) E is a club of the regular cardinal
(b) u = u

: < ), u

and u

= u

(c) for every ES, u

is an unbounded subset of of order-type cf() (and


is a limit ordinal).
By [Sh:420, 1]
{y16}
Claim 0.10. If
+
< and , are regular then some stationary S < :
cf() = belongs to

I[].
By [Sh:108]
{y18}
Claim 0.11. If =
+
, cf() and < [[
<
then S



I[].
(
8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


WHEN FIRST ORDER T HAS LIMIT MODELS SH868 7
1. On superstable not
0
-stable T
We rst note that superstable T tend to have superlimit models.
{nlm.0.0}
Claim 1.1. Assume T is superstable and [T[ + 2
0
. Then T has a superlimit
model of cardinality i T has a saturated model of cardinality i T has a
universal model of cardinality i [D(T)[.
Proof. By [Sh:c, III,5] we know that T is stable in i [D(T)[. Now if
[T[ < [D(T)[ trivially there is no universal model of T of cardinality hence
no saturated model and no superlimit model, etc., recalling 0.7(2). If [D(T)[,
then T is stable in hence has a saturated model of cardinality by [Sh:c, III]
(hence universal) and the class of -saturated models of T is closed under increasing
elementary chains by [Sh:c, III] so we are done.
1.1
The following are the prototypical theories which we shall consider.
{0.0}
Denition 1.2. 1) T
0
= Th(

2, E
0
n
)
n<
when E
0
n
n = n.
2) T
1
= Th(

(
1
), E
1
n
)
n<
where E
1
n
n = n.
3) T
2
= Th(R, <).
Recall
{0.1.1}
Observation 1.3. 0) T

is a countable complete rst order theory for = 0, 1, 2.


1) T
0
is superstable not
0
-stable.
2) T
1
is strictly stable, that is, stable not superstable.
3) T
2
is unstable.
4) T

has elimination of quantiers for = 0, 1, 2.


{0.1.2}
Claim 1.4. It is consistent with ZFC that
1
< 2
0
and some M EC
1
(T
0
) is a
superlimit model.
Proof. By [Sh:100], for notational simplicity we start with V = L.
So T
0
is dened in 1.2(1) and it is the T from Theorem [Sh:100, 1.1] and let S
be the set of (

2)
L
. We dene T

(called T
1
there) as the following theory:

1
(i) for each n the sentence saying E
n
is an equivalence relation with 2
n
equivalence classes, each E
n
equivalence class divided to two by
E
n+1
(ii) the sentences saying that
() for every x, the function z F(x
0
, z) is one-to-one and
() x
0
E
n
F(x
0
, z) for each n <
(iii) E
n
(c

, c

)
if(n=n)
for , S.
In [Sh:100] it is proved that in some forcing
5
extension L
P
of L, P an
2
-c.c. proper
forcing of cardinality
2
, in V = L
P
, the class PC(T

, T
0
) = M
T0
: M is a
-model of T

is categorical in
1
.
However, letting M

be any model from PC(T

, T
0
) of cardinality
1
, it is easy
to see that (in V = L
P
):

2
the following conditions on M are equivalent
(a) M is isomorphic to M

5
We can replace L by any V
0
which satises 2

0
=
1
, 2

1
=
2
.
(
8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


8 SAHARON SHELAH
(b) M PC(T

, T
0
)
(c) () M is a model of T
0
of cardinality
1
() M

can be elementarily embedded into M


() for every a M the set a/E
M
n
: n < has cardinality
1
.
But

3
every model M
1
of T of cardinality
1
has a proper elementary extension
to a model satisfying (c), i.e., (), (), () of
2
above

4
if M

: < ) is an increasing chain of models satisfying (c) of


2
and
<
2
then also M

: < does.
Together we are done.
1.4
Naturally we ask
{0.1.3}
Question 1.5. What occurs to T
0
for >
1
but < 2
0
?
{0.1}
Question 1.6. Does the theory T
2
of linear order consistently have an (
1
,
0
)-
superlimit? (or only strongly limit?) but see 3.
{0.2}
Question 1.7. What is the answer for T when T is countable superstable not

0
-stable and D(T) countable for
1
< 2
0
for
2
< 2
0
?
So by the above for some such T some universe for
1
the answer is yes, there is
a superlimit.
(
8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


WHEN FIRST ORDER T HAS LIMIT MODELS SH868 9
2. A strictly stable consistent example
We now look at models of T
1
(redened below) in cardinality
1
; recall
{sl.1}
Denition 2.1. T
1
= Th(

(
1
), E
n
)
n<
where E
n
= (, ) : ,

(
1
) and
n = n.
{sl.1.3}
Remark 2.2.
(a) Note that T
1
has elimination of quantiers.
(b) If =
n
: n < and
n
=
0
n
, then T
1
has a (,
0
)-superlimit model
in (see 2.15).
{sl.1y}
Denition/Claim 2.3. 1) Any model of T
1
of cardinality is isomorphic to
M
A,h
:= ((, ) : A, < h()), E
n
)
n<
for some A

and h ()
(Car
+
)0 where
1
E
n

2

1
n =
2
n, pedantically E
M
A,h
n
= E
n
[M
A,n
[.
2) We write M
A
for M
A,h
when A is as above and h : A [A[, so constantly [A[
when A is innite.
3) For A

and h as above the model M
A,h
is a model of T i A is non-empty
and ( A)(n < )(
0
A)(n = n (n) ,= (n)).
4) Above M
A,h
has cardinality i h() : A = .
{sl.2}
Denition 2.4. 1) We say that A is a (T
1
, )-witness when
(a) A

has cardinality
(b) if B
1
, B
2


are (T
1
, A)-big (see below) of cardinality then (B
1

>
(, ) is isomorphic to (B
2

>
, ).
2) A set B

is called (T
1
, A)-big when it is (, ) (T
1
, A)-big; see below.
3) B is (, ) (T
1
, A)-big means: B

, [B[ = [A[ = and for every
>

there is an isomorphism f from (

, ) onto ( :

, ) mapping A into
: B.
4) A

(
1
) is
1
-suitable when :
(a) [A[ =
1
(b) for a club of <
1
, A

is everywhere not meagre in the space



, i.e.,
for every
>
the set A

: is a non-meagre subset of

(that is really what is used in [Sh:100]).


{st.3}
Claim 2.5. It is consistent with ZFC that 2
0
>
1
+ there is a (T
1
,
1
)-witness;
moreover every
1
-suitable set is a (T
1
,
1
)-witness.
Proof. By [Sh:100, 2].
2.5
{st.3a}
Remark 2.6. The witness does not give rise to an (
1
,
0
)-limit model as the union
of any reasonable -increasing -chain of members of EC
1
(T
1
) the relevant sets
are meagre.
{sl.4}
Denition 2.7. Let A be a (T
1
, )-witness. We dene K
1
T1,A
as the family of
M = ([M[, <
M
, P
M

such that:
() ([M[, <
M
) is a tree with ( + 1) levels
() P
M

is the -th level; let P


M
<
= P
M
n
: n <
(
8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


10 SAHARON SHELAH
() M is isomorphic to M
1
B
for some B

of cardinality where M
1
B
is
dened by [M
1

[ = (
>
) B, P
M
1
B
n
=
n
, P
M
1
B

= B and <
M
1
B
= [M
1
B
[,
i.e., being an initial segment
() moreover B is such that some f satises:
(a) f :
>
and f(<>) = 0 for simplicity
(b)
>
f() f()
(c) if B then f( n) : n < ) is eventually constant
(d) if
>
then

:

B and m <
f(

( m)) = f() is (T
1
, A)-big
(e) for
>
and n [f(), ) for ordinals < , we have
f(

)) = n.
{sl.5}
Claim 2.8. If A is a (T
1
,
1
)-witness then
(a) K
1
T1,A
,=
(b) any two members of K
1
T1,A
are isomorphic
(c) there is a function F from K
1
T1,A
to itself (up to isomorphism, i.e., (M, F(M))
is dened only up to isomorphism) satisfying M F(M) such that K
1
T1,A
is closed under increasing unions of sequence M
n
: n < ) such that
F(M
n
) M
n+1
.
Proof. Clause (a): Trivial.
Clause (b): By the denition of A is a (T
1
,
1
)-witness and of K
1
T1,A
.
Clause (c):
We choose F such that
if M K
1
A,T1
then M F(M) K
1
A,T1
and for every k < , a P
M
k
the
set b P
F(M)
k+1
: a <
F(M)
b and b / M has cardinality
1
.
Assume M = M
n
: n < where M
n
: n < ) is -increasing, M
n

K
1
A,T1
, F(M
n
) M
n+1
. Clearly M is as required in the beginning of Denition
2.7. Now we dene f : P
M
<
by f(a) = Minn : a M
n
. Pendantically, F is
dened only up to isomorphism.
So we are done.
2.8
{sl.5.1}
Claim 2.9. If A is a (T
1
, )-witness then
(a) K
1
T1,A
,=
(b) any two members of K
1
T1,A
are isomorphic
(c) if M
n
K
1
T1,A
and n < M
n
M
n+1
then M := M
n
: n <
K
1
T1,A
.
Remark 2.10. If we omit clause (b), we can weaken the demand on the set A.
Proof. Assume M = M
n
: n < , M
n
M
n+1
, M
n
K
1
T1,A
and f
n
witnesses
M
n
K
1
T1,A
. Clearly M satises clauses (), (), () from Denition 2.7, we just
have to nd a witness f as in clause () there.
For each n M let n(a) = Minn : a M
n
, clearly if M [= a < b < c then
n(a) n(b) and n(a) = n(c) n(a) = n(b). Let g
n
: M M be: g
d
(a) = b
(
8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


WHEN FIRST ORDER T HAS LIMIT MODELS SH868 11
i b
M
a, b M
n
and b is
M
-maximal under those restrictions; clearly it is
well dened. Now we dene f

n
: M
n
by induction on n < such that
m < n f

m
f

n
, as follows.
If n = 0 let f

n
= f
n
.
If n = m + 1 and a M
n
we let f

n
(a) be f

m
(a) if a M
m
and be (f
n
(a)
f
n
(g
m
(a)))+f

m
(g
m
(a))+1 if a M
n
M
m
. Clearly f := f

n
: n < is a function
from M to , a
M
b f(a) f(b), and for any a M the set b M : a
M
b
and f(b) = f(a) is equal to b M
n(a)
: f
n(a)
(a) = f
n(a)
(b) and a
M
b.
So we are done.
2.9
{sl.5.7}
Denition 2.11. Let A be a (T
1
, )-witness. We dene K
2
T1,A
as in Denition 2.7
but f is constantly zero.
{sl.6}
Claim 2.12. [The Global Axiom of Choice] If A is a (T
1
,
1
)-witness then
(a) K
2
T1,A
,=
(b) any two members of K
2
T1,A
are isomorphic
(c) there is a function F from
+2
(K
2
T0,A
) : <
1
to K
2
T1,A
which satises:
() if

M = M
i
: i + 2) is an -increasing sequence of models
of T then M
+1
F(

M) K
2
T1,A
() the union of any increasing
1
-sequence

M = M

: <
1
)
of members of K
2
T1,A
belongs to K
2
T1,A
when

1
= sup : F(

M ( + 2)) M
+1
) and is a well dened
embedding of M

into M
+2
.
Remark 2.13. Instead of the global axiom of choice, we can restrict the models to
have universe a subset of
+
(or just a set of ordinals).
Proof. Clause (a): Easy.
Clause (b): By the denition.
Clause (c): Let |

: <
1
) be an increasing sequence of subsets of
1
with
union
1
such that <
1
[|


<
|

[ =
1
. Let M

K
2
T1,A
be such that
>
(
1
) [M

[

(
1
) and M

=: M


(|

) belongs to K
2
T1,A
for every
<
1
.
We choose a pair (F, f ) of functions with domain

M :

M an increasing sequence
of members of K
2
T1,A
of length
M
<
1

() F(

M) is an extension of M
i
: i < g(
m
) from K
2
T1,A
() f (

M) is an embedding from M

into F(

M)
() if

M
2
= M

: <
2
),
1
<
2
,

M
1
=

M
2

1
and F(

M
1
) M
1
then
f (

M
1
) f (

M
2
)
() if a F(

M) and n < then for some b M

we have F(M) [= aE
n
(f (

M)(b)).
Now check.
2.12
{sl.7}
Conclusion 2.14. Assume there is a (T
1
,
1
)-witness for the rst-order complete
theory T
1
from 2.1:
1) T
1
has an (
1
,
0
)-strongly limit model.
2) T
1
has an (
1
,
1
)-medium limit model.
(
8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


12 SAHARON SHELAH
3) T
1
has a (
1
,
0
)-superlimit model.
Proof. 1) By 2.8 the reduction of problems on (EC(T
1
), ) to K
1
T1,A
(which is easy)
is exactly as in [Sh:100].
2) By 2.12.
3) Like part (1) using claim 2.9.
2.14
{sl.11}
Claim 2.15. If =
n
: n < and
n
=
0
n
, then T
1
has a (,
0
)-superlimit
model in (see 2.15).
Proof. Let M
n
be the model M
An
where A
n
=

(
n
).
Clearly
()
1
M
n
is a saturated model of T
1
of cardinality
n
()
2
M
n
M
n+1
()
3
M

= M
n
: n < is a special model of T
1
of cardinality .
The main point:
()
4
M

is (,
0
)-superlimit model of T
1
.
[Why? Toward this assume
(a) N
n
is isomorphic to M

say f
n
: M

N
n
is such isomorphic
(b) N
n
N
n+1
for n < .
Let N

= N
n
: n < and we should prove N


= M

, so just N

is a special
model of T
1
of cardinality suce.
Let N

n
= N

(f
n
(M
k
) : k n). Easily N

n
N

n+1
N

and N

n
: n <
= N

and |N

n
| =
n
. So it suces to prove that N

n
is saturated and by
direct inspection shows this.
2.15
(
8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


WHEN FIRST ORDER T HAS LIMIT MODELS SH868 13
3. On non-existence of limit models
Naturally we assume that non-existence of superlimit models for unstable T is
easier to prove. For other versions we need to look more. We rst show that for
[T[ +
1
, if T is unstable then it does not have a superlimit model of cardinality
and if T is unsuperstable, we show this for most cardinals . On proper for
K
or
or K

tr
, see [Sh:c, VII] or [Sh:E59] or hopefully some day in [Sh:e, III].
{nlm.0.1}
Claim 3.1. 1) If T is unstable, [T[ +
1
, then T has no superlimit model of
cardinality .
2) If T is stable not superstable and [T[ +

or =
0
[T[ then T has no
superlimit model of cardinality .
Remark 3.2. We assume some knowledge on EM models for linear orders I and
members of K

tr
as index models, see [Sh:c, VII].
Proof. 1) Towards a contradiction assume M

is a superlimit model of T of cardi-


nality . As T is unstable we can nd m, ( x, y) such that
() ( x, y) L
(T)
linearly orders some innite I
m
M, M [= T so g( x) =
g( y) = m.
We can nd a which is proper for linear orders (see [Sh:c, VII]) and F

( < m)
such that F

T
is a unary function symbol for < m,
T
() and for every
linear order I, EM(I, ) has Skolem functions and its
T
-reduct EM
(T)
(I, ) is a
model of T of cardinality [T[ +[I[ and () is of cardinality [T[ +
0
and a
s
: s I)
is the Skeleton of EM(I, ), that is, it is an indiscernible sequence in EM(I, ) and
EM(I, ) is the Skolem hull of a
s
: s I, and letting a
s
= F

(a
s
) : < m) in
EM(I, ) we have EM
(T)
(I, ) [= [ a
s
, a
t
]
if(s<t)
for s, t I.
Next we can nd
n
(for n < ) such that:
(a)
n
is proper for linear order and
0
=
(b) EM
()
(I,
n
) EM
()
(I,
n+1
) for every linear order I and n < ;
moreover
(b)
+
(
n
) (
n
) and EM(I,
n
) EM
(n)
(I,
n+1
) for every
n < and linear order I
(c) if [I[ n then EM
()
(I,
n
) = EM
()
(I,
n+1
) and
EM
(T)
(I,
n
)

= M

(d) [(
n
)[ = .
This is easy. Let

be the limit of
n
: n < ), i.e. (

) = (
n
) : n <
and EM
(
k
)
(I,

) = EM
(
k
)
(I,
n
) : n [k, ). So as M

is a superlimit
model, for any linear order I of cardinality , EM
(T)
(I,

) is the direct limit of


EM
(T)
(J,

) : J I nite), each isomorphic to M

, so as we have assumed that


M

is a superlimit model it follows that EM


(T)
(I,

) is isomorphic to M

. But
by [Sh:300, III] or [Sh:E59] which may be [Sh:e, III] there are 2

many pairwise
non-isomorphic models of this form varying I on the linear orders of cardinality ,
contradiction.
2) First assume =
0
. Let
T
be countable such that T

= T L() is not
superstable. Clearly if M

is (,
0
)-limit model then M

is not
1
-saturated.
But if = cf() [
1
, ] and M

is a (, )-limit then M

is
1
-saturated,
contradiction.
(
8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


14 SAHARON SHELAH
The case [T[ +

is more complicated (the assumption

is to enable
us to use [Sh:460] or see [Sh:829] for a simpler proof; we can use weaker but less
transparent assumptions; maybe 2
0
suces).
As T is stable not superstable for some

:

1
for any there are M and a
,
:

and < ) such that
(a) M is a model of T
(b) I

= a
,
: < M is an indiscernible set (and < <
a
,
,= a
,
)
(c)

=
n
: n < ) and
n
L
(T)
innite
(d) for ,

we have Av
n
(M, I

) = Av
n
(M, I

) i n = n.
Hence by [Sh:c, VIII], or see [Sh:E59] assuming M

is a universal model of T of
cardinality :

2.1
there is such that
(a) is proper for K

tr
,
T
(), [()[ = [T[ +
0
(b) for I

, EM
()
(I, ) is a model of T and I J EM(I, )
EM(J, )
(c) for some two-place function symbol F if for I K

tr
and P
I

, I a
subtree of

for transparency we let I
I,
= F(a

, a

) : I then
I
I,
: P
I

) are as in
1
(b), (d).
Also

2.2
if
1
satises (a),(b),(c) of
2.1
and M is a universal model of T then there
is

2
satisfying (a),(b),(c) of
2
and
1

2
see
2.3
(a) and for every
nitely generated J K

tr
, see
2.3
(b) below, there is M

= M such that
EM
(T)
, (J, ) M

EM
(T)
(J,

2
)

2.3
(a) we say
1

2
when (
1
) (
2
) and J K

tr
EM(J,
1
)
EM
(1)
(J,
2
)
(b) we say J I is nitely generated if it has the form

: < n
: for some n, we have P
I
n
and <
I

for some

0
, . . . ,
n1
P
I

2.4
if M

EC

(T) is superlimit (or just weakly S-limit, S


+
stationary)
then there is as in
2.1
above such that EM
(T)
(J, )

= M

for every
nitely generated J K

tr

2.5
we x as in
2.4
for M

EC

(T) superlimit.
Hence (mainly by clause (b) of
2.1
and
2.4
as in the proof of part (1))

3
if I K

tr
has cardinality then EM
()
(I, ) is isomorphic to M

.
Now by [Sh:460], we can nd regular uncountable <

such that =
[]
, i.e.,
for every set A of cardinality there is T
A
[A]

= B A : [B[ of
cardinality such that any B []

is the union of < members of T


A
.
Let S = < : cf() =
0
and =

: S) be such that

an increasing
sequence of length with limit .
For a model M of T let OB

(M) = a : a = a

,
: W and < ), W S
and in M they are as in
1
(b), (d).
(
8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


WHEN FIRST ORDER T HAS LIMIT MODELS SH868 15
For a OB

(M) let W[ a] be W as above and let
( a, M) =

: there is an indiscernible set
I = a

: < in M such that for every n


for some W[ a], n =

n and
Av
n
(M, I) = Av
n
(M, a

,
: < ).
Clearly

4
(a) if M N then OB

(M) OB

(N)
(b) if M N and a OB

(M) then ( a, M) ( a, N).
Now by the choice of it should be clear that

5
if M [= T is of cardinality then we can nd an elementary extension
N of M of cardinality such that for every a OB

(M) with W[ a] a
stationary subset of , for some stationary W

W[ a] the set [ a, N]
includes

: (n)( W

)( n =

n), (moreover we can even


nd

< and W

W for <

satisfying W[ a] = W

: <

6
we nd M EC

(T) isomorphic to M

such that for every a OB



(M)
with W[ a] a stationary subset of , we can nd a stationary subset W

of
W[ a] such that the set [ a, M] includes

: (n)( W

)( n =

n).
[Why? We choose (M
i
, N
i
) for i <
+
such that
(a) M
i
EC

(T) is -increasing continuous


(a) M
i+1
is isomorphic to M

(a) M
i
N
i
M
i+1
(a) (M
i
, N
i
) are like (M, N) in
5
.
Now M = M
i
: i <
+
is as required.
Now the model M is isomorphic to M

as M

is superlimit.]
Now the model from
6
is not isomorphic to M

= EM
(T)
(
>

:
S, ) where is from
2.1
. But M

= M

by
3
.
Together we are done.
3.1
The following claim says in particular that if some not unreasonable pcf conjec-
tures holds, the conclusion holds for every 2
0
.
{nlm.0.1g}
Claim 3.3. Assume T is stable not superstable, [T[ and = cf() >
0
.
1) T has no (, )-superlimit model provided that =
0
and = U
D
() :=
Min[T[ : T []

and for every f : for some u T we have < :


f() u D
+
, where D is a normal lter on to which < :cf() =
0

belongs.
2) Similarly if 2
0
and letting J
0
= u : [u[
0
, J
1
= u : u S

0
non-stationary we have = U
J1,J0
() := Min[T[ : T []
0
, if u J
1
, f :
(u) then for some countable innite w (u) and v T, Rang(fw) v.
Proof. Like 3.1.
(
8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


16 SAHARON SHELAH
{n9}
Claim 3.4. 1) Assume T is unstable and [T[ +

. Then for at most one


regular does T have a weakly (, )-limit model and even a weakly (, S)-limit
model for some stationary S S

.
2) Assume T is unsuperstable and [T[ +

(
2
) and
1
=
0
<
2
= cf(
2
).
Then T has no model which is a weak (, S)-limit where S and S S

is
stationary for = 1, 2.
Proof. 1) Assume
1
,=
2
form a counterexample. Let <

be regular large
enough such that =
[]
, see proof of 3.1(2) and /
1
,
2
. Let m, ( x, y) be
as in the proof of 3.1
() if M EC

(T) then there is N such that


(a) N EC

(T)
(b) M N
(c) if a = a
i
: i < )

(
m
M) for < then for some | []

for every
uniform ultralter D on to which | belongs there is a
D

n
N such
that tp( a
D
, N, N) = Av(D, a, M) = ( x, c) : ( x, z) L(
T
), c
g( z)
M and < : N [= [ a
i
, c] D.
Similarly
for every function F with domain

M :

M an -increasing sequence of
models of T of length <
+
each with universe
+
such that M
i

F(

M) for i < g(

M) and F(

M) has universe
+
there is a sequence
M

: <
+
) obeying F such that: for every <
+
and a

(
m
(M

))
for < , there is | []

such that for every ultralter D on to


which | belongs, for every (,
+
) there is a
D,

m
(M
+1
) realizing
Av(D, a, M

) in M
+1
.
Hence
for lM

: <
+
) as in
1
for every limit <
+
of conality ,= for
every a = a
i
: i < )

(
m
(M

)), there is | []

such that for every


ultralter D on to which | belongs, there is a sequence

: < cf())
cf()
(
m
(M

)) such that for every ( x, z) L(


T
) and c
g( z)
(M

) for every
< cf() large enough, M

[= [

, c] i ( x, c) Av(D, a, M

).
The rest should be clear.
2) Combine the above and the proof of 3.1(2).
3.4
(
8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


WHEN FIRST ORDER T HAS LIMIT MODELS SH868 17
4. Private Appendix
More on unstable T
We next show that an unstable T has no models which are quite weakly limit
(this is close to older proofs of many non-isomorphic models). If T has the strict
order property, things are easier.
{nlm.1}
Claim 4.1. Assume that T is a complete unstable rst order theory. If [T[+
1
then T has no model M which is simultaneously weakly (,

)-limit for = 1, 2
where
1
,=
2
are regular .
This will be proved after some claims
{nlm1.1}
Claim 4.2. If M is a weakly (, )-limit,
1
<
2
, ,=
1
, ,=
2
and ,
1
,
2
are regular and in Th(M), ( x, y) as in 3.1 and its proof then

1,2,
M
there are no sequences a

( <
1
, <
2
) from M satisfying M [=
[ a
1
, a
2
] [ a
2
,

b
2
] [

b
2
,

b
1
] when
1

2
<
1
,
1

2
<
2
and
for every c
m
M for some truth value t and

<
1
,

<
2
we have:
[

, ) M [= [ a

, c]
t
, [

,
2
) M [= [ c,

]
t
.
Proof. For any M model of T of cardinality there is M
+
, M M
+
, |M
+
| =
such that concerning
1,2,
M
+
, it holds if a

are from M; see [Sh:c, VIII,0] but


we prove it for completeness. Without loss of generality, (T) is countable and

1
<
2
and we prove it by induction on .
So if M
i
: i < ) is -increasing each M
i
a model of T of cardinality and
(M
i
, M
i+1
) is like (M, M
+
) above, then easily M

= M
i
: i < which is a model
of T of cardinality satises
1,2,
M
because for any a

( <
1
),

( <
2
) as
above for some () < we have [I
1
[ =
1
, [I
2
[ =
2
where I
1
:= <
1
: a


M
()
and I
2
:= sup <
2
:

b

M
()
.
4.2
{nlm.1.2}
Claim 4.3. If M is a weakly (, )-limit model; = cf() , T = Th(M) has
cardinality and ( x, y) as in 3.1 for T and
1
= cf(
1
) ,
1
,= then
1,,
M
fails.
Proof. We can nd

M such that
(a)

M = M
i
: i ), a -increasing sequence of models of T of cardinality
(b) a

: <
1
) are such that M
0
[= [ a
1
, a
2
] for
1
<
2
<
1
(c) b
i+1
M
i+1
realizes in M
i+1
over M
i
the type ( x, c) : ( x, z) L()
and <
1
: ( a

, c) D for some uniform ultralter D on


1
.
Easily a

: <
1
),

: < ); exemplify the failure.


4.3
{nlm.1.2a}
Discussion 4.4. So by 4.2 + 4.3 we have: no M EC

(T) can be (, )-superlimit


for three s.
{nlm.1.3}
Claim 4.5. Assume

1
T,
[T[
1
= cf() and M is a model of T of cardinality and is a
weakly (, )-limit model

2

T
is countable and

3
[M[ = a
i
: i < and let A
i
:= a
j
: j < i for i < .
(
8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


18 SAHARON SHELAH
Then

1
for some club of limit ordinals < for every c
n
M we have

,c
if tp
L()
( c, A

, M) is nitely satisable in A

, then for some ultral-


ter D on
n
(A

) we have Av(D, M) includes tp


L()
( c, A

, M) and its
restriction to A

is realized in M for every <

2
if ( x, y) L(
T
) is as in 3.1 for T then in addition in M there is an
indiscernible sequence a = a
i
: i < ) such that
() M [= [ a

, a

] i
() the sequence converges, i.e., for any formula ( x, z) L(
T
) and c
>
M for some truth value t for every < large enough, M [=
[ a

, c]
t
() moreover we can nd

b = b
t
: t I
2

) where on the linear order I


2

see below, such that



b
(,)
= a

,

b is an indiscernible sequence and for
each limit < , for some ultralter D

on containing the co-bounded


subsets of , for every (, ), b
(,)
realizes Av
D
(A

, a

: < ))
where
() I
2

= , (
1
,
1
) <
I
2

(
2
,
2
)
1
<
1
(
1
=
2

1
>
2
).
{nlm.1.4}
Claim 4.6. Assume
(a)
1
+[T[ = cf(), = cf() <
(b) M is a weakly (, )-limit model
(c)

A = A

: < ) satises [A
i
[ < , A
i
increases continuous with
union [M[
(d) =

: < ) is increasing continuous with union


T
, [

[ < .
Then

for a club of < , for every



b M we can nd c

: < ) realizing
tp
L(

)
(

b, A

, M) such that
if

d M and ( x, z) L(

) and for some nite q tp


L(0)
(

b, A

, M)
we have ( e A

)( e realizes q e satises ( x,

d)) then for every
< large enough we have M [= [ c

,

d].
Proof. We dene F such that
if M
2
= F(M
1
) and M
1
, M
2
are models of T of cardinality , A

: < )
represents M

, then for a club of < , for every



b M
1
for some c M
2
we have:
for every ( x, z) L(

) and

d M
1
and for some nite q tp
L(0)
(

b, A
1
0
, M
1
)
we have ( e A
1

)( e realizes q e satises ( x,

d)).
We choose by induction on i < , M
i
,

A
i
,

D
i

b,
:

b M) such that:
(1) M
i
EC
T
() is -increasing
(2)

A
i
= A

: < ) is a model of M
i
(
8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


WHEN FIRST ORDER T HAS LIMIT MODELS SH868 19
(3) if

b M
i
and tp
L(

)
(

b, A

) is nitely satisable in A

then D
i

b,
is an
ultralter on
g(

b)
(A

) such that
tp
L(

)
(

b, A
i

, M
i
) = Av(A

, D
i

b,
)
(4) M
i

= M
i+1
(5) for every

b, as in (3) there is some c
i
b,

M
i+1
realizing Av(M
i
, D
i

b,
).
The rest should be clear.
4.6
{nlm.1.5}
Proof. Proof of 4.1 when =
1
:
Let =
1
, I
2

be as in
2
of 4.5 and also ( x, y), a

: < ),

b
(,i)
: (, i) I
2

)
and a as there.
So for some thin enough club E of for every E the demand there holds. In
particular (M

) A

and for every ( x, z) L(

) and e
g( z)
(A

)
there are t = t
( x, e)
and
( x, e)
< such that for every [
( x, e)
, ) we have
M [= [ a

, e]
t
and in particular, M [= [ a

, e] [ a

, e].
As the number of such ( x, e) is [[ < then for some
2

< we have:
()
1
if
1
( x, c)
< and
2

< i < then


M [= (

b
(,i)
, e) ( a

, e) ( a

, e).
We shall show that there are no c

: < ) as in 4.6 for , a

. How? If < , i <


then M [= [ a

, a
(,i)
] so by ()
1
and the property of the c

: < ) we know that


()
2
for each i < for every < large enough
M [= [ c

b
(,i)
].
By the property of the

b
(,i)
: i < ).
()
3
for each < for i < large enough M [=

b
(,i)
, c

).
As ( x, y) ( y, x) and cf() ,= cf() we get a contradiction.
4.1
References
[Mirar] Dzamonja Mirna, Club guessing and the universal models, On pcf (Ban, Alberta, 2004)
(Matthew Foreman, ed.), to appear.
[Sh:c] Saharon Shelah, Classication theory and the number of nonisomorphic models, Studies in
Logic and the Foundations of Mathematics, vol. 92, North-Holland Publishing Co., Amsterdam,
xxxiv+705 pp, 1990.
[Sh:e] , Nonstructure theory, vol. accepted, Oxford University Press.
[Sh:h] , Classication Theory for Abstract Elementary Classes, Studies in Logic: Mathe-
matical logic and foundations, vol. 18, College Publications, 2009.
[Sh:E59] , General non-structure theory and constructing from linear orders.
[Sh:88] , Classication of nonelementary classes. II. Abstract elementary classes, Classi-
cation theory (Chicago, IL, 1985), Lecture Notes in Mathematics, vol. 1292, Springer, Berlin,
1987, Proceedings of the USAIsrael Conference on Classication Theory, Chicago, December
1985; ed. Baldwin, J.T., pp. 419497.
[Sh:88r] , Abstract elementary classes near
1
, Chapter I. 0705.4137. 0705.4137.
[Sh:100] , Independence results, The Journal of Symbolic Logic 45 (1980), 563573.
(
8
6
8
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
2
6







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
2
7


20 SAHARON SHELAH
[Sh:108] , On successors of singular cardinals, Logic Colloquium 78 (Mons, 1978), Stud.
Logic Foundations Math, vol. 97, North-Holland, Amsterdam-New York, 1979, pp. 357380.
[Sh:300] , Universal classes, Classication theory (Chicago, IL, 1985), Lecture Notes in
Mathematics, vol. 1292, Springer, Berlin, 1987, Proceedings of the USAIsrael Conference on
Classication Theory, Chicago, December 1985; ed. Baldwin, J.T., pp. 264418.
[KjSh:409] Menachem Kojman and Saharon Shelah, Non-existence of Universal Orders in Many
Cardinals, Journal of Symbolic Logic 57 (1992), 875891, math.LO/9209201.
[Sh:420] Saharon Shelah, Advances in Cardinal Arithmetic, Finite and Innite Combinatorics in
Sets and Logic, Kluwer Academic Publishers, 1993, N.W. Sauer et al (eds.). 0708.1979, pp. 355
383.
[Sh:460] , The Generalized Continuum Hypothesis revisited, Israel Journal of Mathematics
116 (2000), 285321, math.LO/9809200.
[Sh:829] , More on the Revised GCH and the Black Box, Annals of Pure and Applied Logic
140 (2006), 133160, math.LO/0406482.
[Sh:877] , Dependent T and Existence of limit models, Tbilisi Mathematical Journal sub-
mitted, math.LO/0609636.
[Sh:900] , Dependent theories and the generic pair conjecture, Communications in Con-
temporary Mathematics submitted, math.LO/0702292.
[Sh:906] , No limit model in inaccessibles, Proceedings of the Makkaifest accepted,
0705.4131.
[Sh:950] , A dependent dream and recounting types.
[Sh:F1054] , On weakly limit models.
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

Potrebbero piacerti anche