Professional Documents
Culture Documents
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
Categoricity of Theories in L
, when
is a measurable cardinal. Part 1
Saharon Shelah
1
and
Oren Kolman
Institute of Mathematics
Hebrew University
Jerusalem, Israel
Abstract
We assume a theory T in the logic L
, _
F
) when one of the extension is nice (2.1). We
prove that if T is categorical in the regular > [T[ + , then (/
<
, _
F
) has the amal-
gamation property. For this we show that nice extension (in /
<
) preserve being non
amalgamation basis. We also start investigating (in 2.5) the connection between extend-
ing the linear order I and the model EM(I): I
nice
J EM(I) _
nice
EM(J); and give
sucient condition for I
nice
J (in 2.6). From this we get in K
K = Mod(T) then
for a club of i <
+
, M
i
_
nice
M
j
: j <
+
. We dene nice models (essentially, every
reasonable extension is nice). Def (3.2) show a variant is equivalent def (3.4), implies being
amalgamation base (3.5), prove that in K
in K
is (, )-saturated over M
0
. We show existence (and iniqueness). We connect
this to more usual saturation and prove that (, )-saturation implies niceness (in 4.10).]
5 The amalgamation property for /
<
:
[After preliminaries we prove that for ( and [T[ + of course) every member
of K
can be extended to one with many nice submodels, this is done by induction on
using the niceness of (
1
,
1
)-saturated models. Lastly we conclude that every M K
<
is nice hence K
<
has the amalgamation property.]
3
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
0 INTRODUCTION:
The main result of this paper is a proof of the following theorem:
Theorem. Suppose that T is a theory in a fragment of L
where is a measurable
cardinal. If T is categorical in the cardinal > + [T[, then /
<
, the class of models
of T of power strictly less than (but + [T[), has the amalgamation property (see
Denition 1.3 (1)(2)).
The interest of this theorem stems in part from its connection with the study of
categoricity spectra. For a theory T in a logic L let us dene Cat(T), the categoricity
spectrum of T, to be the collection of those cardinals in which T is categorical. In the
1950s Los conjectured that if T is a countable theory in rst-order logic, then Cat(T)
contains every uncountable cardinal or no uncountable cardinal. This conjecture, based
on the example of algebraically closed elds of xed characteristic, was veried by Morley
[M], who proved that if a countable rst-order theory is categorical in some uncountable
cardinal, then it is categorical in every uncountable cardinal. Following advances made by
Rowbottom [Ro], Ressayre [Re] and Shelah [Sh2], Shelah [Sh31] proved the Los conjecture
for uncountable rst-order theories: if T is a rst-order theory categorical in some cardinal
> [T[ +
0
, then T is categorical in every cardinal > [T[ +
0
. It is natural to ask
whether analogous results hold for theories in logics other than rst-order logic. Perhaps
the best-known extensions of rst-order logic are the innitary logics L
. As regards
theories in L
, then the
Los conjecture holds for . Furthermore, under some set-theoretic assumptions (weaker
than the Generalized Continuum Hypothesis) if is a sentence in L
which is categorical
in
n
for every natural number n (or even just if is a sentence in L
with at least
one uncountable model not having too many models in each
n
), then is excellent. Now
[Sh300], [Sh-h] try to develop classication theory in some non elementary classes. we
cannot expect much for L
for >
0
. Shelah conjectured that if is a sentence in
L
1
, then is categorical in every >
1
. (Recall that
4
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
the Hanf number of L
is
1
, so if is a sentence in L
1
, then has a model in every power
1
. See [K].) There were some
who asked why so tardy a beginning. Recent work of Hart and Shelah [Sh323] showed
that for every natural number k greater than 1 there is a sentence
k
in L
which
is categorical in the cardinals
0
. . . ,
k1
, but which has many models of power for
every cardinal 2
k1
. The general conjecture for L
: (1)
if is a successor cardinal and > ((
)
+
where
)
+
), (2) if >
+1
(
), then T
is categorical in every cardinal of the form
with divisible by (2
)
+
(i.e. for some
ordinal > 0 = (2
)
+
(ordinal multiplication)). In proving theorems of this kind,
one has recourse to the amalgamation property which makes possible the construction
of analogues of saturated models. In turn these are of major importance in categoricity
arguments. The amalgamation property holds for theories in rst-order logic [CK] and
in L
fails the
Tarski-Vaught property for unions of chains of length (where as
L
and
L
coincide). However it is
not known in general for theories in L
or L
when is measurable.
A sequel to this paper under preparation tries to provide a characterization of Cat(T)
at least parallel to that in [MaSh285] and we hope to deal with the corresponding classica-
tion theory later. This division of labor both respects historical precedent and is suggested
by the increasing complexity of the material. Another sequel deals with abstract elemen-
tary classes (in the sense of [Sh88]) (see [Sh472], [Sh394] respectively). On more work see
[Sh576], [Sh600].
5
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
The paper is divided into ve sections. Section 1 is preliminary and notational. In
section 2 it is shown that if T is categorical in the regular cardinal > +[T[, then /
<
has the amalgamation property. Section 3 deals with weakly universal models, section 4
with (, )-saturated and
-saturated models. In section 5 the amalgamation property for
/
<
is established.
All the results in this paper (other than those explicitly credited) are due to Saharon
Shelah.
1 PRELIMINARIES:
To start things o in this section, let us x notation, provide basic denitions and
well-known facts, and formulate our working assumptions.
The working assumptions in force throughout the paper are these.
Assumption 1: The cardinal is an uncountable measurable cardinal, and so there is a
-complete nonprincipal ultralter on .
Assumption 2: The theory T is a theory in the innitary logic L
.
From these assumptions follow certain facts, of which the most important are these.
Fact 1: For each model M of T, -complete ultralter D over I and suitable set G of
equivalence relations on I I (set 1.7.4) the limit ultrapower Op(M) = Op(M, I, D, G) is
a model of T.
Fact 2: For each linear order I = (I, ) there exists a generalized Ehrenfeucht-Mostowski
model EM(I) of T.
The remainder of this section provides more detailed explanations and references.
Relevant set-theoretic and model-theoretic information on measurable cardinals can
be found in [J][CK][D]. L denotes a language, i.e. a set of nitary relation and func-
tion symbols, including equality. [L[ is the cardinality of the language L. For a cardinal
, L
: <
and L
: <
) of variables of length
: <
). So if ( x) is a
formula in L
, so there is a fragment T of L
= M K : [[M[[ = , K
<
=
<
K
, K
, K
[,)
=
<
K
.
We write f : M
F
N (abbreviated f : M N) to mean that f is an T- elementary
embedding (briey, an embedding) of M into N, i.e. f is a function with domain [M[ into
[N[ such that for every T - formula ( x), and a
<
[M[ with lg( a) = lg( x), M [ a]
i N [f( a)], where if a = a
i
: i < n), then f( a) := f(a
i
) : i < n). In the special case
where an embedding f is a set-inclusion (so that [M[ [N[), we write M
F
N (briey
M N) instead of f : M
F
N and we say that M is an T - elementary submodel of N,
or N is an T-elementary extension of M.
(I,
I
), (J,
J
) . . . are partial orders; we will not bother to subscript the order
relation unless really necessary; we write I for (I, ). (I, ) is directed i for every i
1
and i
2
in I, there is i I such that i
1
i and i
2
i. (I, <)
, <
) where I
= I and s <
t i t < s.
7
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
A set M
i
: i I) of models indexed by I is a
F
- directed system i (I, ) is a
directed partial order and for i j in I, M
i
F
M
j
. The union
iI
M
i
of a
F
- directed
system M
i
: i I) of L - structures is an L - structure. In fact, more is true.
Fact 1.1: (Tarski-Vaught property)
(1) The union of a
F
- directed system M
i
: i I) of models of T is a model of T, and
for every j I, M
j
F
iI
M
i
.
(2) If M is a xed model of T such that for every i I there is f
i
: M
i
F
M and for
all i j in I, f
i
f
j
, then
iI
f
i
:
iI
M
i
F
M. In particular, if M
i
F
M for every
i I, then
iI
M
i
F
M. Let be an ordinal. A
F
- chain of models of length
is a sequence M
F
M
. The
chain is continuous if for every limit ordinal < , M
=
<
M
.
Fact 1.2: (Downward Loewenheim Skolem Property): Suppose that M is a model of
T, A [M[ and max( + [T[, [A[) [[M[[. Then there is a model N such that
A [N[, [[N[[ = and N
F
M.
Finally, > +[T[ usually denotes a power in which T is categorical.
Now we turn from the rather standard model-theoretic background to the more specic
concepts which are central in our investigation.
Definition 1.3:
(1) Suppose that < is a binary relation on a class K of models. / = K, <) has the
amalgamation property (AP) i for every M, M
1
, M
2
K, if f
i
is an isomorphism
from M onto rng(f
i
) and rng(f
i
) < M
i
for i = 1, 2, then there exist N K and
isomorphisms g
i
from M
i
onto rng(g
i
) for i = 1, 2 such that rng(g
i
) < N and g
1
f
1
=
g
2
f
2
. The model N is called an amalgam of M
1
, M
2
over M with respect to f
1
, f
2
.
(2) An L-structure M is an amalgamation base (a.b.) for / = K, <) i M K and when-
ever for i = 1, 2, M
i
K and f
i
is an isomorphism from M onto rng(f
i
), rng(f
i
) < M
i
,
then there exist N K and isomorphisms g
i
(i = 1, 2) from M
i
onto rng(g
i
) such that
rng(g
i
) < N and g
1
f
1
= g
2
f
2
.
So / = K, <) has AP i every model in K is an a.b. for /.
Example 1.3A: Suppose that T is a theory in rst-order logic having an innite model.
8
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
Dene, for M, N in the class K
|T|+
0
of models of T of power at most [T[ +
0
, M < N
i there is an embedding of M into an elementary submodel of N. Then /
|T|+
0
=
K
|T|+
0
, <) has AP [CK].
Example 1.3B: Suppose that T is a theory in L
and T is a fragment of L
containing
T with [T[ < [T[
+
+ . Let < be the binary relation
F
dened on the class K of all
models of T. M K is an a.b. for / i whenever for i = 1, 2, M
i
K and f
i
is an
F
-
elementary embedding of M into M
i
, there exist N K and T - elementary embeddings
g
i
(i = 1, 2) of M
i
into N such that g
1
f
1
= g
2
f
2
.
Definition 1.4:
Suppose that < is a binary relation on a class K of models. Let be a cardinal.
M K
and isomorphisms f
i
from M into M
i
such that
(A) rng(f
i
) < M
i
(i = 1, 2),
(B) there is no amalgam N K
of M
1
, M
2
over M with respect to f
1
, f
2
.
Observation 1.5:
Suppose that T, T and < are as in 1.3B and +[T[ < . Note that if there is an
amalgam N
of M
1
, M
2
over M (for M
1
, M
2
, M in K
of M
1
, M
2
over M.
Indiscernibles and Ehrenfeucht - Mostowski Structures
The basic results on generalized Ehrenfeucht-Mostowski models can be found in [Sh-a]
or [Sh-c Ch VII]. We recall here some notation . Let I be a class of models which we call
the index models. Denote the members of I by I, J . . . For I I we say that a
s
: s I)
is indiscernible in M i for every s,
t
<
I realizing the same atomic type in I, a
s
and
a
t
realize the same type in M (where a
s
0
,...,s
n
= a
s
0
. . . a
s
n
). If L L
are languages
and is a function with domain including tp
at
( s, , I) : s
<
I and I I, we let
EM
(I, ) be an L
-model generated by
sI
a
s
such that tp
at
( a
s
, , M) =
tp
at
( s, , I)
.
We say that is proper for I if for every I I, EM
(I, ) is well-dened.
9
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
Let EM(I, ) be the L-reduct of EM
(I, ) by EM
.
Definition and Proposition 1.6:
Suppose that T is a theory, T T where T is a fragment of L
.
(A) There are a (canonically constructed) nitary language L
sk
and a universal theory
T
sk
in L
such that
(0) L L
sk
, [L
sk
[ [T[ +
0
.
(1) The L - reduct of any L
sk
- model of T
sk
is a model of T.
(2) Whenever N
sk
is an L
sk
- model of T
sk
and M
sk
is a substructure of N
sk
, then
M
sk
L
F
N
sk
L.
(3) Any L - model of T can be expanded to an L
sk
- model of T
sk
.
(4) If M
F
N, then there are L
sk
- expansions M
sk
, N
sk
of M, N respectively such
that M
sk
is a substructure of N
sk
and N
sk
is a model of T
sk
.
(5) To any T - formula ( x), those corresponds a quantier-free formula
qf
( x) of
(L
sk
)
such that
T
sk
x((( x)
qf
( x)).
Limit ultrapowers, iterated ultrapowers and nice extensions.
An important technique we shall use in studying the categoricity spectrum of a theory
in L
E, where G
is a lter on I I, then
D/G
M =
D/G
M.
Definition 1.7.2:
Let M be an L-structure, Y, <) a linear order and, for each y Y , let D
y
be an
ultralter on a non-empty set I
y
. Write H =
yY
I
y
. Let
yY
D
y
= s H : there are
y
1
< . . . < y
n
in Y such that
(1) for all i, j H, if i y
1
, . . . , y
n
= j y
1
, . . . , y
n
then i s i j s;
(2) i(y
1
), . . . , i(y
n
)) : i s D
y
1
. . . D
y
n
.
Write E =
yY
D
y
. The iterated ultrapower
E
[M[ of the set [M[ with respect to D
y
: y
Y ) is the set f/E : f : H [M[ and for some nite Z
f
Y for all i, j H, if i Z
f
= j
Z
f
, then f(i) = f(j). The iterated ultrapower
E
M of the L-structure M with respect to
D
y
: y Y ) is the L-structure whose universe is the set
E
[M[; for each n-ary predicate
symbol R of L, R
E
M
(f
1
/E, . . . , f
n
/E) i i H : R
M
(f
1
(i), . . . , f
n
(i)) E; for each n-
ary function symbol F of L, F
E
M
(f
1
/E, . . . , f
n
/E) = F
M
(f
1
(i), . . . , f
n
(i)) : i H)/E.
The canonical map d : M
E
M is dened as usual by d(a) = a : i H)/E.
Remark 1.7.3:
(1) Every ultrapower is a limit ultrapower: take G = P(I I) and note that
D
M =
D/G
M.
11
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
(2) Every iterated ultrapower is a limit ultrapower. [Why? let the iterated ultrapower
be dened by Y, <) and (I
y
, D
y
) : y Y ) (see Denition 1.7.2). For Z [Y ]
<
, let
A
Z
= (i, j) H H : i Z = j Z. Note that A
Z
: Z [Y ]
<
has the nite
intersection property and hence can be extended to a lter G on HH. Now for any
model M we have
E
M
=
D/G
M for every lter D over H extending E under the
map f/E f/D.]
Definition 1.7.4:
Suppose that M is an L-structure, D is an ultralter on a non-empty set I, and G is
a suitable set of equivalence relations on I, i.e.
(i) if e G and e
G;
(ii) G is closed under nite intersections;
(iii) if e G, then D/e = A I/e :
xA
x D is a -complete ultralter on I/e which,
for simplicity, has cardinality .
Then Op(M, I, D, G) is the limit ultrapower
D/
G
M where
G is the lter on I I generated
by G. One abbreviates Op(M, I, D, G) by Op(M), and one writes f
Op
for the canonical
map d : M Op(M).
Note that
Observation / Convention 1.7.4B:
1) For any L-structure N, f
Op
is an L
. Given
-complete ultralters D
1
on I
1
, D
2
on I
2
and suitable lters G
1
on I
1
I
1
, G
2
on I
2
I
2
respectively, there exist a -complete ultralter D on a set I and a lter G on I I such
that
Op(M, I, D, G) = Op(Op(M, I
1
, D
1
, G
1
), I
2
, D
2
, G
2
)
and (D, G, I) is -complete. Also iterated ultrapower (along any linear order) which each
iterand being ultrapower by -complete ultralter, gives a suitable triple (in fact, even
iteration of suitable limit ultrapowers is a suitable ultrapower).
Definition 1.8:
Suppose that K is a class of L-structures and < is a binary relation on K. For
M, N K, write f : M
nice
<
N to mean
(1) f is an isomorphism from M into N and rng(f) < N
(2) there are a set I, an ultralter D on I, a suitable set G of equivalence relations on
I (so Denition 1.7.4 clause (i), (ii), (iii) holds), and an isomorphism g from N into
Op(M, I, D, G) such that rng(g) < Op(M, I, D, G) and gf = f
Op
, where f
Op
is the
canonical embedding of M into Op(M, I, D, G). f is called a < - nice embedding of
M into N. Of course one writes f : M
nice
N and says that f is a nice embedding of
M into N when < is clear from the context.
Example 1.9.1:
Consider T, T and / = K, <) as set up in 1.3B. In this case f : M
nice
<
N holds i
f : M
F
N and for some suitable I, D, G) and some g : N
F
Op(M, I, D, G), gf = f
Op
.
Abusing notation one writes M
nice
N to mean that there are f, g and Op such that
f : M <
nice
<
N using g and Op. IF NOT SAID OTHERWISE, < is <
F
. We may
also write M _
nice
N, and for linear orders we use I
nice
J.
Example 1.9.2:
13
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
Let LO be the class of linear orders and let (I,
I
) < (J,
J
) mean that (I,
I
)
(J,
J
), i.e. (I,
I
) is a suborder of (J,
J
). If f : (I,
I
)
nice
<
(J,
J
), then for some Op,
identifying isomorphic orders, one has (I,
I
) (J,
J
) Op(I,
I
).
Observation 1.10:
Suppose that T, T and / are as in 1.3B and 1.9.1. Suppose further that M <
nice
N
and M _
F
M
_
F
N for M, M
, N K. Then M <
nice
M
.
Proof: For some f, g and Op, f : M
F
N, g : N
F
Op(M) and gf = f
Op
. Now
g : M
F
Op(M) (since M
_
F
N) and gf = f
Op
so that M <
nice
M
.
Observation 1.11:
Suppose that M
i
: i ) is a continuous increasing chain and for each i < ,
M
i
<
nice
M
i+1
. Then for every i < , M
i
<
nice
M
.
Proof of 1.11: Like the proof of 1.7.3(2). For each i < , there is a (I
i
, D
i
, G
i
) as in
Defenition 1.7.4 which witness M
i
nice
M
i+1
. Let I
def
=
i<
I
i
, G = e : e I I and
for some n < and
1
< . . .
n
< , and e
1
G
1
, . . . , e
n
G
n
we have: for every
x, y I, such that (x(
), y(
)) e
1
, . . . , e
n
G
n
, e
equivalence relation on I
for = 1, . . . , n and A (D
1
/e
1
) (D
n
/e
n
), then
the set x I : x(
1
)/e
1
, . . . , x(
n
)/e
n
) A belongs to D. We leave the rest to the
reader.
Claim 1.12: For every model M and +[T[ +|M| there is N such that M _
F
nice
N,
M ,= N and |N| = .
Proof: As is measurable.
2 The Amalgamation Property for regular categoricity:
The main aim in this section is to show that if T is categorical in the regular cardinal
> +[T[, then /
<
= K
<
, _
F
) has the amalgamation property (AP) (Denition 1.3
(1)). Categoricity is not presumed if not required.
14
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
Lemma 2.1.
Suppose that + [T[ , M, M
1
, M
2
K
, f
1
: M
nice
M
1
, f
2
: M
F
M
2
.
Then there is an amalgam N K
of M
1
, M
2
over M with respect to f
1
, f
2
.
Moreover, there are g
: M
F
N for = 1, 2 such that g
1
f
1
= g
2
f
2
and rng(g
2
) =
rng(g
1
f
1
).
Proof: There are g and Op such that g : M
1
F
Op(M), gf
1
= f
Op
. f
2
induces an
T - elementary embedding f
2
of Op(M) into Op(M
2
) such that f
2
f
Op
= f
Op
f
2
. Let
g
1
= f
2
g and g
2
= f
Op
M
2
. By 1.2 one nds N K
2
gf
1
= f
2
f
0
p
= f
0
p
f
2
= g
2
f
2
. The last phrase in the lemma is easy by properties
of Op.
Lemma 2.2.
Suppose that M K
and f
i
: M
F
M
i
(i = 1, 2) there is no amalgam of M
1
, M
2
over M w.r.t. f
1
, f
2
. Let f
i
be
the T - elementary embedding from Op(M) into Op(M
i
) dened by f
i
(note that f
i
f
Op
=
f
Op
f
i
, i = 1, 2). Choose N
i
of power [[N[[ such that M
i
rng(f
i
g) N
i
F
Op(M
i
).
Note that f
i
f : N
F
N
i
. It suces to show that there is no amalgam of N
1
, N
2
over N
w.r.t. f
1
g, f
2
g.
Well, suppose that one could nd an amalgam N
and h
i
: N
i
F
N
, i = 1, 2,
with h
1
(f
1
g) = h
2
(f
2
g). Using 1.2 choose M
, [[M
[[ , M
_
F
N
, rng(h
1
f
Op
M
1
) rng(h
2
f
Op
M
2
) [M
[. Set g
i
= h
i
f
Op
M
i
, for i = 1, 2, and note that
g
1
f
1
= h
1
f
Op
f
1
= h
1
f
1
f
Op
= h
1
f
1
gf = h
2
f
2
gf = h
2
f
2
f
Op
=
= h
2
f
Op
f
2
= g
2
f
2
.
In other words, M
is an amalgam of M
1
, M
2
over M w.r.t. f
1
, f
2
- contradiction. It
follows that N is a |N|-c.a.b.
15
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
Corollary 2.3:
Suppose that + [T[ < . If M K
such that
() M _
F
M
K
<
, if M _
F
M
_
F
M
, then M
is a [[M
[[ - c.a.b.
Proof:
As |M| , for some appropriate Op one has [[Op(M)[[ , and by 1.2 one nds
M
such that M M
_
F
Op(M). Lets check that M
K
<
, M _
F
M
_
F
M
; so M _
nice
M
since M
_
F
Op(M); hence by 2.2, M
is a
[[M
[[ - c.a.b
Theorem 2.4.
Suppose that T is - categorical, = cf() > + [T[. If /
<
fails AP, then there
is N
=
i<
N
i
(2) for every i < , N
i
nice
N
i+1
(and so N
i
nice
N
).
Proof: /
<
fails AP, so for some < and M K
, M is a - c.a.b. By 2.2
and 1.12 wlog M K
=
i<
N
j
i
. So |N
K
<
and M _ M
_ M
, then M
is a
|M
|-c.a.b.
Since T is - categorical, there is an isomorphism g of N
onto M
. Let N = g
1
(M)
and M
i
= g(N
i
) for i < . [[N[[ = < cf() = , so there is i
0
< such that N N
j
i
0
.
16
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
In fact N
j
i
0
is a [[N
j
i
0
[[ - c.a.b. [Otherwise, consider M
j
i
0
. Since M _
F
M
j
i
0
_
F
M
and |M
j
i
0
| < , M
j
i
0
is a |M
j
i
0
| - c.a.b., so there are f
: M
j
i
0
F
M
, ( = 1, 2), with
no amalgam of M
1
, M
2
over M
j
i
0
w.r.t. f
1
, f
2
. If N
j
i
0
is not a [[N
j
i
0
[[ - c.a.b., then
one can nd an amalgam N
+
/
||N
j
i
0
||
of M
1
, M
2
over N
j
i
0
w.r.t. f
1
g, f
2
g such that
h
: M
F
N
+
and h
1
(f
1
g) = h
2
(f
2
g); so h
1
f
1
= h
2
f
2
and N
+
is thus an amalgam
of M
1
, M
2
over M
j
i
0
w.r.t. f
1
, f
2
, |N
+
| |N
j
i
0
| = [[M
j
i
0
[[ - contradiction.] This
contradicts the choice of N
j
i
0
. So the claim is correct.
It follows that for each j (j
0
, ) there are N
1
j
, N
2
j
in K
<
and f
: N
j
F
N
j
such
that no amalgam of N
1
j
, N
2
j
over N
j
w.r.t. f
1
, f
2
exists. By 2.2 for some 1, 2, N
j
nice
N
j+1
. So by the inductive choice of N
j+1
: j < ), j (j
0
, )(N
j
nice
N
j+1
). Taking
N
=
j
0
<j<
N
j
, one completes the proof. (Of course for j
0
< j < , N
j
nice
N
: if
N
j
_
F
N
_
F
Op(N
j
), then by 1.10 N
j
_
nice
N
j+1
- contradiction).
Theorem 2.5. Suppose that (I, <
I
), (J, <
J
) are linear orders and I is a suborder of J. If
(I, <
I
)
nice
(J, <
J
), then EM(I) _
nice
EM(J).
Proof: WLOG for some cardinal , ultralter D on and suitable set G, a lter on
, (I, <
I
) (J, <
J
) Op((I, <
I
), , D, G) = Op(I, <), and [Op(I, <)[ = f/D : f
I, eq(f) G where eq(f) = (i, j) : f(i) = f(j). So for each t J, there exists
f
t
I such that t = f
t
/D. Note that if t I, then f
t
/D = f
Op
(t) so that WLOG for all
i < , f
t
(i) = t. Dene a map h from EM(J) into Op(EM(I)) as follows. An element of
EM(J) has the form
EM
(J)
(x
t
1
, . . . , x
t
n
)
where t
1
, . . . , t
n
J, an L-term. Dene, for t J, g
t
EM(I) by g
t
(i) = x
f
t
(i)
.
Note that f
t
(i) I, so that x
f
t
(i)
EM(I) and so g
t
/D Op(EM(I)). Let h(
EM
(J)
(x
t
1
, . . . , x
t
n
)) =
Op(EM
(I))
(g
t
1
/D, . . . , g
t
n
/D) which is an element in Op(EM(I)). The
reader is invited to check that h is an T - elementary embedding of EM(J) into Op(EM(I)).
So EM(I) _
F
EM(J).
Finally note that if =
EM
(I)
(x
t
1
, . . . , x
t
n
) EM(I), t
1
, . . . , t
n
I, then h( ) =
Op(EM
(I))
(g
t
1
/D,...,g
t
n
/D
) =
Op(EM
(I))
(x
t
1
: i < )/D, . . . , x
t
n
: i < )/D) =
f
Op
(
EM
(I)
(x
t
1
, . . . , x
t
n
)) = f
Op
( ). Thus EM(I) _
nice
EM(J).
17
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
Criterion 2.6:
Suppose that (I, <) is a suborder of the linear order (J, <).
If () for every t J I
() cf((I, <) s I : (J, <) s < t) =
or
() cf((I, <)
s I : (J, <)
s <
t) =
then (I, <)
nice
(J, <),
[Notation: (I, <)
, <
) where I
= I and (I
, <
) s <
t
i (I, <) t < s.]
Proof:
Let us list some general facts which facilitate the proof.
Fact (A): Let denote the linear order (, <) where < is the usual order . If
J
1
= + J
0
, then
nice
J
1
(+ is addition of linear orders in which all elements in the rst
order precede those in the second).
Fact (B): If (I, <), is unbounded in (I, <) and J
1
= I +J
0
, then I
nice
J
1
.
Fact (C): If I
nice
J, then I +J
1
nice
J + J
1
.
Fact (D): I
nice
J i (J <)
nice
(I, <)
.
Fact (E): If I
nice
I
+1
, then I
nice
I
.
Now using these facts, let us prove the criterion. Dene an equivalence relation E on
J I as follows: tEs i t and s dene the same Dedekind cut in (I, <). Let t
: <
be a set of representatives of the E-equivalence classes. For each , dene
I
= J t : t I
<
tEt
so I
0
= I, I
= J and I
nice
I
+1
for each < .
18
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
Fix < . Now t
= .
Let
I
a
= t I
: t < t
I
b
= t I
+1
: t I
a
tEt
I
c
= t I
: t > t
.
Note that I
= I
a
+I
c
and I
+1
= I
b
+I
c
nice
I
b
. Identifying isomorphic orders and using (), one has that is unbounded
in I
a
and I
b
= I
a
+ (I
b
I
a
) so by Fact (B), I
a
nice
I
b
as required.
Of the ve facts, we prove (A), (B) and (E) as (C) and (D) are obvious.
Proof of Fact (A): Since is measurable, there is a -complete uniform ultralter
D on [see J]. For every linear order J
0
(or J
0
) there is Op
I,D
(), the iteration of I
ultrapowers ()
0
), giving the required embedding (use
1.7.5).
Proof of Fact (B): Since I and using Fact (A), we know that there is an operation
Op such that the following diagram commutes:
I
id
Canonical
Canonical
Op()
Natural
Op(I)
id
Fact(A)
+J
0
id
I + J
0
Chasing through the diagram, we obtain the required embedding.
Proof of Fact (E): Apply 1.11 to the chain I
: ).
Fact 2.7:
19
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
Suppose that . There exist a linear order (I, <
I
) of power and a sequence
A
i
I : i ) of pairwise disjoint subsets of I, each of power such that I =
i
A
i
and
() if X 1, then I
iX
A
i
nice
I.
Proof: Let I = ( 1) and dene <
I
on I : (i
1
,
1
) <
I
(i
2
,
2
) i i
1
< i
2
or
(i
1
= i
2
and
1
>
2
). For each i , let A
i
= i . Check () of 2.6: suppose that
X + 1. Write I
X
= I
iX
A
i
. To show that I
X
nice
I, one deploys Criterion
2.6. Consider t I I
X
, say t = (i, ) (note that < and i < , since X) and
i / X. Let j = min(X i); note that j is well-dened, since X i, and j ,= i. For
every < , one has t <
I
(j, ) and (j, ) I
X
. Also if s I
X
and t <
I
s, then for some
< (j, ) <
I
s. Thus (j, ) : < ) is a conal sequence in (I
X
s I : t <
I
s)
.
By the criterion, I
X
nice
I.
Theorem 2.8. Suppose that = cf() < . Then EM() _
nice
EM().
Proof: By Fact (B) of 2.6, one has that
nice
; so by 2.5, EM() _
nice
EM().
Now lets turn to the main theorem of this section.
Theorem 2.9. Suppose that T is categorical in the regular cardinal > + [T[. Then
/
<
has the amalgamation property.
Proof: Suppose that /
<
fails AP. Note that |EM()| = . Apply 2.4 to nd M
and M
i
: i < ) satisfying 2.4 (1) and 2.4 (2). Since T is -categorical, M
= EM(),
so WLOG EM() =
i<
M
i
. C = i < : M
i
= EM(i) is a club of . Choose
C, cf() = . By 2.8, EM() _
nice
EM(), so M
_
nice
M
nice
M
- contradiction.
Theorem 2.10. Suppose that T is categorical in > +[T[. Then:
(1) T has a model M of power such that if N _
F
M and |N| < , then there exists N
such that
() N _
F
N
_
F
M
() |N
| = |N| + +[T[
20
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
() N
_
nice
M.
(2) T has a model M of power and an expansion M
+
of M by at most +[T[ functions
such that if N
+
M
+
, then N
+
L _
nice
M.
Proof: Let I, A
i
: i )) be as in 2.7. Let M = EM(I). Suppose that N _
M, |N| < . Then there exists J I, [J[ < such that N EM(J) hence N _
F
EM(J) _
F
EM(I) . So there is X + 1, X, [X[ < such that J
iX
A
i
. Note
that [
iX
A
i
[ [X[ < . Now N
= EM(I
iX
A
i
) is as required, since I
iX
A
i
nice
I
and so by 2.5 EM(I
iX
A
i
) _
nice
EM(I). This proves (1).
(2) We expand M = EM(I) as follows
(a) by all functions of EM
(I)
(b) by the unary functions f
b
1) from I such that b =
b
(x
t(b,0)
, x
t(b,1)
, . . . , x
t(b,n
b
1)
) (it is not unique,
but we can choose one; really if we choose it with n
b
minimal it is almost unique).
We let
f
(b) =
x
t((b,)
if < n
b
b if n
b
(c) by unary functions g
, g
I
] then
g
(x
t
) = x
s
,
<
g
(x
s
) = x
t
(more formally g
(x
(i,)
) = x
(i,+)
and g
(x
(i,)
) =
x
(i,)
) in the other cases g
(b) = b, g
(b) = b.
(d) by individual constants c
= x
(,)
for < .
Now suppose N
+
is a submodel of M
+
and N its L-reduct. Let J
def
= t I : x
t
N,
now J is a subset of I of cardinality [[N[[ as for t ,= s from J, x
t
,= x
s
. Also if b N by
clause (b), x
t(b,)
N hence b EM(J); on the other hand if b EM(J) then by clause
(b) we have b N; so we can conclude N = EM(J). So far this holds for any linear order
I.
By clause (c) J =
iX
A
i
for some X + 1, and by clause (d), X.
Now EM(J) _
nice
EM(I) ,= M by 2.7.
21
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
3 Towards removing the assumption of regularity from the existence of
universal extensions:
In 2 we showed that /
<
has the amalgamation property when T is categorical in
the regular cardinal > + [T[. We now study the situation in which is not assumed
to be regular.
Our problem is that while we know that most submodels of N K
sit well in N
(see 2.10(2)) and that there are quite many N K
<
which are amalgamation bases, our
diculty is to get those things together: constructing N K
as
i<
N
i
, N
i
K
<
means
N has _
F
-submodels not included in any N
i
.
Theorem 3.1. Suppose that T is categorical in and +[T[ < . If M
i
K
: i <
+
) is an increasing continuous
F
-chain, then
i <
+
: M
i
_
nice
j<
+
M
j
D
+.
Remark 3.1A:
(1) We cannot use 2.10(1) as possibly has conality < +[T[.
(2) Recall that D
+
is the closed unbounded lter on
+
.
Proof: Write M
+
=
i<
+
M
i
. Choose an operation Op such that for all
i <
+
|Op(M
i
)| . Let M
i
= Op(M
i
). Applying 1.2 for non-limit ordinals, 1.1
for limit ordinals, one nds inductively an increasing continuous
F
-chain N
i
: i
+
)
such that for i <
+
, M
i
_ N
i
_ M
i
, |N
i
| = , N
+
=
i<
+
N
i
. Note that |N
+
| =
+
=
Since T is -categorical, N
+
= EM(I) where 2.7 furnishes I of power . By 2.10
(2), there is an expansion N
+
+
of N
+
by at most +[T[ functions such that if A [N
+
+
[
is closed under the functions of N
+
+
, then (N
+
+
L) A _
nice
N
+
.
Choose a set A
i
and an ordinal j
i
, by induction on i <
+
, satisfying
(1) A
i
[N
+
[, [A
i
[ ; A
i
: i <
+
) is continuous increasing;
(2) j
i
: i <
+
) is continuous increasing;
(3) A
i
is closed under the functions of N
+
+
;
(4) A
i
[N
j
i+1
[;
(5) [M
i
[ A
i+1
.
22
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
This is possible: for zero or limit ordinals unions work; for i + 1 choose j
i+1
to satisfy (2)
and (4), and A
i+1
to satisfy (1), (3) and (5).
By (2), C = i <
+
: i is a limit ordinal and j
i
= i is a club of
+
i.e. C D
+
.
Fix i C. Note that [M
i
[ A
i
and A
i
[N
i
[ (since [M
i
[ =
j<i
[M
j
[
j<i
A
j+1
=
A
i
=
i
<i
A
i
i
<i
[N
j
i
+1
[ = N
j
i
= N
i
(using (5), (1), (4), (2) and j
i
= i)) and so
M
i
_
F
(N
+
+
L) A
i
_
F
N
i
_
F
M
i
= Op(M
i
), so that M
i
_
nice
(N
+
+
L) A
i
. However
by (3) and the choice of N
+
and N
+
+
one has also that (N
+
+
L) A
i
_
nice
N
+
. So by
transitivity of _
nice
, one obtains M
i
_
nice
N
+
.
Finally remark that M
+
_ N
+
since M
i
_
nice
N
i
_
F
N
+
for every i <
+
.
Hence C i <
+
: M
i
_
nice
M
+
D
+
.
Definition 3.2:
Suppose that [ + [T[, ) and M K
. M is nice i whenever M _
F
N K
,
then M _
nice
N. (The analogous T - elementary embedding denition runs: M is nice i
whenever f : M
F
N K
then f : M
nice
N.)
Theorem 3.3.
Suppose that T is categorical in and M K
such that M _
F
N and N is nice.
Proof: Suppose otherwise. Well dene a continuous increasing
F
- chain M
i
K
: i <
+
) such that for j <
+
()
j
M
j
nice
M
j+1
.
For i = 0, put M
0
= M; if i is a limit ordinal, put M
i
=
j<i
M
j
; if i = j + 1,
then, since 3.3 is assumed to fail, M
j+1
exists as required in ()
j
(otherwise M
i
works as
N in 3.3). But now M
i
: i <
+
) yields a contradiction to 3.1, since C = i <
+
:
M
i
_
nice
j<
+
M
j
D
+
by 3.1 so that choosing j from C one has M
j
_
nice
M
j+1
by 1.10,
contradicting ()
j
.
Theorem 3.4.
23
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
Suppose that T is categorical in and [ + [T[, ). If M K
is nice and
f : M
F
N K
, then f : M
nice
N.
Proof: Choosing an appropriate Op and using 1.2 one nds N
1
such that N _
F
N
1
and
|N
1
| = . Find M
1
_
nice
N
1
by 2.10(2) such that rng(f) [M
1
[, |M
1
| = . So M
1
_
F
N
1
and so rng(f) _
F
M
1
. M is nice, so f : M
nice
M
1
. Now M
1
_
nice
N
1
, so f : M
nice
N
1
.
So there are Op and g : N
1
F
Op(M) satisfying gf = f
Op
. Since N _
F
N
1
it follows that
f : M
nice
N as required.
Corollary 3.5: Suppose that M K
i.e. if f
i
: M
F
M
i
, M
i
K
of M
1
, M
2
over M w.r.t. f
1
, f
2
.
Proof: : By 3.4 f
i
: M
nice
M
i
(i = 1, 2). Hence by 2.1 there is an amalgam N K
of M
1
, M
2
over M w.r.t. f
1
, f
2
.
Definition 3.6: Suppose that [ +[T[, ) and is a cardinal.
(1) A model M K
2
K
2
F
M
2
such that f M
1
is the identity. (The embedding version
runs: there exists h : M
1
F
M
2
and whenever g : M
1
F
M
2
K
2
F
M
2
such that fg = h.) M
2
is universal over M
1
(M
1
_
univ
M
2
) i M
2
is |M
2
|
- universal over M
1
.
(3) M
2
is -universal over M
1
in M i M
2
_
F
M
1
_
F
M, [[M
1
[[ and whenever
M
2
K
and M
1
_
F
M
2
_
F
M, then there exists an T - elementary embedding
f : M
2
M
2
such that f M
1
is the identity. M
2
is universal over M
1
in M i M
2
is |M
2
| - universal over M
1
in M.
(4) M
2
is weakly -universal over M
1
(written M
1
wu
M
2
) i M
1
_
F
M
2
K
and whenever M
2
F
M
2
K
2
F
M
2
such that f M
1
is the identity. (The embedding version is: there exists
h : M
1
F
M
2
and whenever g : M
2
F
M
2
K
2
F
M
2
such
24
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
that h = fgh (written h : M
1
wu
M
2
).). M
2
is weakly universal over M
1
(M
1
_
wu
M
2
)
i M
2
is |M
2
| - weakly universal over M
1
.
(note: M
2
is -universal over M
1
does not neccessarily implies M
2
is weakly -
universal over M
1
as possibly |M
2
| > )
Remark 3.6A:
(5) In K
<
, if M
1
is an a.b., then weak universality over M
1
is equivalent to universality
over M
1
.
Proof: Suppose that h : M
1
wu
M
2
and g : M
1
F
M
2
K
M
2
. Since M
1
is an a.b.
there exist a model N and h
: M
2
F
N, g
: M
2
F
N satisfying h
h = g
g. By 1.2 WLOG
|N| = |M
2
|. Since M
2
is weakly universal over M
1
, there exists h
: N
F
M
2
, h = h
h.
Let f = h
: M
2
M
2
, and note that fg M
1
= h
g = h
h = h, so that M
2
is
universal over M
1
.
Remark 3.6B:
(6) For any model M, universality over M implies weak universality over M.
Lemma 3.7: Suppose that T is categorical in , [ + [T[, ). If M K
and
M _
F
N K
such that
(1) M _
F
M
+
_
F
N,
(2) M
+
is universal over M in N.
Proof: Now choose I such that
()
into J
0
J
1
which is the identity on J
0
.
Essentially the construction follows Laver [L] and [Sh 220, Appendix]; but for our
present purpose let I = (
>
, <
ex
); given and J
0
we can increase J
0
so wlog J
0
=
>
A,
A , [A[ = . Dene an equivalence relation E on I J
0
: E ( J
0
)( <
ex
<
ex
), easily it has equivalence classes, so let
i
: i < i
be a set of
representative each of minimal length, so
(lg
i
1) J
0
,
i
(lg
i
1) A.
25
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
Let J
1
= I
:
>
and i < i
, so clearly J
0
J
1
I, [J
1
= .
Suppose J
0
J I, [J[ , and we should nd the required embedding h. As before
wlog J =
>
B, [B[ = and A B. Now h J
0
= id
J
0
so it is enough to dene
h (J
1
(
i
/E)), hence it is enough to embed J
1
(
i
/E) into
:
>
(under
<
ex
).
Let = otpB, so it is enough to show (
<
, <
ex
) can be embedded into
>
, where
of course [[ . This is proved by induction on .
Since T is - categorical and EM(I) is a model of T of power , there is an isomor-
phism g from EM(I) onto N. It follows from () that M
+
= g
EM(J) K
satises (1)
and (2). (Analogues of (1) and (2) are checked in more detail in the course of the proof of
3.11.)
Lemma 3.8: Suppose that T is categorical in , [ + [T[, ), and M
i
K
: i <
+
), N
i
K
: i <
+
) are continuous
F
- chains such that for every i <
+
M
i
_
F
N
i
.
Then there exists i() <
+
such that (i(),
+
) C = i <
+
: M
i+1
can be T -
elementarily embedded into N
i
over M
0
.
Proof: Apply 3.7 for M
0
K
and N
+ =
i<
+
N
i
K
(noting that M
0
_
F
N
0
_
F
N
+)
to nd M
+
K
such that M
0
_
F
M
+
_
F
N
+
and M
+
is universal over M
0
in N
+
.
For some i() <
+
, M
+
N
i()
and so M
+
_
F
N
i()
. If i (i(),
+
), then
M
i+1
K
and M
0
_
F
M
i+1
_
F
N
i+1
_
F
N
. Then
there exists M
+
K
such that
() M _
F
M
+
and M
+
is nice
() M
+
is weakly universal over M.
Proof: Dene by induction on i <
+
continuous
F
- chains M
i
K
: i <
+
), N
i
K
: i <
+
) such that
26
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
(0) M
0
= M ;
(1) M
i
_
F
N
i
;
(2) if ()
i
holds, then M
i+1
cannot be T - elementarily embedded into N
i
over M
0
,
where ()
i
is the statement: there are M
and N
such that M
i
_
F
M
,
N
i
_
F
N
, M
_
F
N
and M
, M
i
_
F
M
_
nice
N
i
. Let
M
i+1
= M
, N
i+1
= N
i
. If ()
i
does hold for M
, N
, let N
i+1
= N
, M
_
F
M
_
nice
N
; now let M
i+1
= M
such that M
i+1
_
F
M
_
F
N
i+1
and M
such that M
_
F
M
+
and M
+
is nice.
So M
+
satises (). It remains to show that M
+
is weakly universal over M. Suppose
not and let g : M
+
F
M
where M
over M, hence in N
i+1
over
M. M
i+1
_
F
M
and by (3) M
i+1
_
nice
N
i+1
K
of M
, N
i+1
. The existence of M
, N
implies that ()
i+1
holds since M
cannot
be T - elementarily embedded into N
i+1
over M
0
, hence M
i+2
cannot be T - elementarily
embedded into N
i+1
in contradiction to the choice of i as by 3.7 i + 1 is in C.
Corollary 3.10: If T is categorical in , [ +[T[, ) and M K
is an a.b. (e.g. M
is nice - see 2.1), then there exists M
+
K
such that
() M _
F
M
+
and M
+
is nice
() M
+
is universal over M.
Proof: 3.9 and 3.6 (5).
Corollary 3.11: Suppose that T is categorical in and [ + [T[, ). Then there is
a nice universal model M K
.
27
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
Proof: By 3.3 it suces to nd a universal model of power , noting that universality is
preserved under T - elementary extensions in the same power.
As in the proof of 3.7, there is a linear order (I, <
I
) of power and J I , [J[ = ,
such that ():
(J
I) (if [J
into J).
Claim: EM(J) K
is universal.
Proof: EM(J) is a model of power since max([J[, +[T[) and = [J[ |EM(J)|
EM(J) is universal. Suppose that N K
, N _
F
M, so that by - categoricity of T, M
= EM(I). There is a subjective
T - elementary embedding h : N
F
N
_
F
EM(I) and there exists J
I , [J
[
|N
EM(J
)
into EM(J). Let f = gh, then f : N
F
EM(J) is as required.
Theorem 3.12.
Suppose that T is categorical in , [ + [T[, ), N K
<
is nice, M K
and
M _
nice
N. Then M is nice.
Proof: Let B K
, M _
F
B. Show that M _
nice
B. Well, since M _
nice
N and M _
F
B,
by 2.1 there exists an amalgam M
K
<
of N, B over M. WLOG by 1.5 |M
[[ = [[N[[.
N is nice, hence N _
nice
M
. Since M _
nice
N, it follows by 1.7.5 that M _
nice
M
. Since
M _
F
B _
F
M
.
Definition 4.1: Suppose that is a limit ordinal, 1 [ +[T[, ).
(1) An L - structure M is (, ) - saturated i
(a) |M| = ;
28
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
(b) there exists a continuous
F
- chain M
i
K
. For i = j +1,
note that M
j
is an a.b. by 3.5 (since M
j
is nice), hence by 3.10 there exists a nice model
M
i
K
, M
j
_
F
M
i
, M
i
universal over M
j
. For limit i, let M
i
=
j<i
M
j
. Note that by the
inductive hypothesis (3) on for i < , since M
i
is (, i) - saturated, M
i
is nice. Thus
M =
i<
M
i
is (, ) - saturated (witnessed by M
i
: i < )). Note that M is universal,
since M
i
: i < ) is continuous and M
0
is universal.
Ad (2). As is a limit ordinal standard back-and-forth argument shows that if M
and N are (, ) - saturated models, then M and N are isomorphic.
Ad (3). By the uniqueness (i.e. by Ad(2)) it suce to prove that some (, ) saturated
model is nice. Suppose that M is (, ) - saturated. Well show that M is nice.
If cf() < , then M is also (, cf ()) - saturated and hence by the inductive
hypothesis (3) on for cf(), M is nice. So well assume that cf() = . Choose a
continuous
F
- chain M
i
K
: i <
+
) such that: M
0
is nice and universal (possible
by 3.11); if M
i
is nice, then M
i+1
K
, M
i
_
F
M
i+1
and
M
i
nice
M
i+1
. By 3.1 and 1.10 there is a club C of
+
such that if i C, then M
i
_
nice
M
i+1
.
29
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
So by the choice of M
i
: i <
+
), if i C, then M
i
is nice. Choose i C, i = sup(i C),
cf(i) = . It suces to show that M
i
is (, ) -saturated (for then by (2) M
i
is isomorphic
to M and so M is nice). Choose a continuous increasing sequence
: < ) C such
that i = sup
<
. Of course
M
0
is universal (since M
0
is universal and M
0
_
F
M
0
), M
+1
is universal over M
since M
+1
is universal over M
and M
_
F
M
+1
_
F
M
+1
. Also M
is nice for
each < since
C. Hence M
i
is (, ) - saturated.
Remark 4.3:
Remember that by 3.12, if T is categorical in , [ + [T[, ), N K
<
is nice,
M K
and M _
nice
N. Then M is nice.
Theorem 4.4.
Suppose that T is categorical in , +[T[ <
+
< . If M
i
K
: i <
+
) is a
continuous
F
-chain of nice models such that M
i+1
is universal over M
i
for i <
+
, then
i<
+
M
i
is (
+
,
+
)-saturated.
Proof: Write M =
i<
+
M
i
. Note that if M
i
K
: i <
+
) is any other continuous
F
-chain of nice models such that M
i+1
is universal over M
i
, then
i<
+
M
= M (use
again the back and forth argument).
By 4.2 there exists a (
+
,
+
)-saturated model N which is unique and nice. In partic-
ular [[N[[ =
+
and there exists a continuous _
F
-chain N
i
K
+
: i <
+
) such that (i)
N
0
is nice and universal, (ii) N
i+1
is universal over N
i
, (iii) N
i
is nice, (iv) N =
i<
+
N
i
.
It suces to prove that M and N are isomorphic models.
WLOG [N[ =
+
. By 1.2 C
1
= <
+
: N _
F
N contains a club of
+
. By 3.1
there exists a club C
2
of
+
such that for every C
2
, N _
nice
N. Since [N
i
[ : i <
+
is
a continuous increasing sequence of subsets of
+
, it follows that C
3
= <
+
: [N
[
is a club of
+
. Hence there is a club C
4
of
+
such that C
4
C
1
C
2
C
3
[,
+
).
Note that for C
4
one has N _
nice
N, [N [ = [N
[ and N
_
F
N, so that
N _
F
N
_
F
N and so by 1.10 N _
nice
N
. N
: C
4
) is a continuous increasing
F
-chain, N
+
and N K
.
30
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
By 3.12 N is nice since N
_ N. [Why? since
N _
F
B
(in fact N _
nice
B
) and N _
nice
N
+ of B
, N
over N . Let f
: B
F
A
be a witness. But N
+1
is universal
over N
(by (ii)), so A
can be
F
-elementarily embedded into N
+1
over N
(say by g
),
hence B
can be
F
-elementarily embedded into N (using g
).]
Let C
5
= C
4
: if C
4
, then [B
[ . Note that C
5
is a club of
+
since
[[B
[[ = . [For C
4
, let E
= (sup [B
[,
+
) C
4
, E
=
+
for , C
4
and let E be the
diagonal intersection of E
: <
+
), i.e. E = <
+
: ( < )( E
). Note that E
is club of
+
and C
5
EC
4
which is a club of
+
.] Thus N : C
5
) is a continuous
F
-chain of nice models, each of power . If
1
C
5
and
2
= min(C
5
(
1
+ 1)),
then N
1
_
F
B
1
_
F
N
2
. Hence N
2
is universal over N
1
(since B
1
is
universal over N
1
). Let
i
: i <
+
enumerate C
5
and set M
i
= N
i
. Note that
N =
i<
+
M
i
. Then M
i
K
: i <
+
) is a continuous
F
-chain of nice models, M
i+1
is universal over M
i
. Therefore N and M are isomorphic (as said in the beginning of the
proof), as required.
Notation 4.5.
= : =
i
: i < ) is a (strictly) continuous increasing sequence of cardinals,
+[T[ <
0
, <
0
(a limit ordinal),
i
and
: sup
i
< .
Remark 4.6.
Let = sup rng() for
. Then is singular, since cf() <
0
.
Definition 4.7.
Let
. A model M is -saturated i there is a continuous
F
-chain M
i
K
i
:
i < ) such that M =
i<
M
i
, M
i
is nice and M
i+1
is
i+1
-universal over M
i
.
Definition 4.8.
Suppose that . Pr() holds i every -saturated model is nice.
Remark 4.9:
31
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
1) If
1
,
2
, rng(
1
) rng(
2
), sup rng(
1
) = sup rng(
1
), and M is
2
-saturated,
then M is
1
-saturated.
2) For
and Pr(
) whenever
and for every limit ordinal < lg(), Pr( ). Then Pr().
Proof: By 4.9.1 and the uniqueness of -saturated models (4.9.2), WLOG one may as-
sume that otp() = cf(sup rng()). Now by 4.6 (cf())
+
< , so by [Sh 420 1.5 + 1.2(1)]
there exists S, C
) cf();
() if C
, then S and C
= C
;
() C
: S
1
. We shall construct the required model by induction,
using C
: S). Remember
=
, M
(for <
+
, < cf()), and N
(only when
S
C
) such that:
(A1) M
;
(A2) M
: <
+
) is a continuous increasing
F
-chain of models;
(A3) M
+1
is nice, and if M
,_
nice
M
+1
;
(A4) M
,= M
+1
;
(A5) M
+1
is weakly universal over M
;
(B1) M
=
<cf()
M
, [[M
[[ =
;
if S
1
, C
, C
, < , then
(B2) N
_
F
M
;
(B3) [[N
[[ =
otp(C
)
;
(B4) ( < otp(C
))(M
);
32
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
(B5) N
is nice;
(B6) N
is
otp(C
)
-universal over N
.
There are now two tasks at hand. First of all, we shall explain how to construct these
entities (THE CONSTRUCTION). Then we shall use them to build a nice -saturated
model (PROVING Pr()). From the uniqueness of -saturated models it will thus follow
that Pr() holds.
THE CONSTRUCTION:
Case (i): = 0. Choose M
0
K
and M
0
K
.
Case(ii): is a limit ordinal. Let M
=
<
M
and choose M
since C
such that M
_
F
M
and if possible M
,_
nice
M
; wlog M
. If , S, then dene
things as above, taking into account (A3). The denitions of M
, M
present no special
diculties. Now suppose that S. The problematic entity to dene is N
.
If C
= , choose for N
)
) of M
.
If C
=
C
. Note that N
is nice. [If C
has a last
element
, then N
= N
which is nice; if C
=
C
is otp(C
)), so N
is
nice.] Also N
_
F
M
. If otp(C
= N
and M
= M
, so
we have nished, so assume otp(C
, [A[
j
< , then there exist a nice M
+
K
, M _
F
M
+
, and nice
models N
, N
+
K
j
, A N
_
F
N
+
_
F
M
+
and N
+
is universal over N
.
Why is this enough? Use the Lemma with M = M
and A = N
<otp(C
)
C
to
nd N
, N
+
, M
+
and choose N
+
, M
+
as N
, M
respectively.
Why () holds? The proof of () is easy as M
is nice.
PROVING Pr():
33
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
For S
1
, consider N
: C
). For , C
<otp(C
)
M
. Therefore M
(M
=
<cf()
M
=
<cf()
M
( S
1
); for < cf(), choose C
, < , < ;
so M
and M
).
Thus for every C
, M
hence M
=
C
(remember
= sup C
as S
1
). If C
, then N
_
F
M
= M
by continuity. So M
=
C
hence N
: C
) exemplies M
is
)). So M
is -saturated S
1
and is stationary, so, applying
3.1, there exists <
+
such that M
is -saturated and M
_
nice
<
+
M
. Hence by 1.10
M
_
nice
M
+1
and so, since M
+1
is nice (A3), M
, = sup
i
. Every -saturated model is (, cf())-saturated.
Proof: Let M
: <
+
) be as in the proof of 4.10. By 3.1 there exists a club C of
+
such that for every C, M
_
nice
<
M
is nice. So
if , C and < , then M
is a universal extension of M
is (, cf())-saturated. Choose S
1
C and sup( C) = .
So M
i
, and +[T[
0
< . Then there exist a linear order I of power and
a continuous increasing sequence I
i
: i < cf()) of linear orders such that
34
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
(1) +[T[ [I
i
[
i
;
(2)
i<cf()
I
i
= I;
(3) every t I
i+1
I
i
denes a Dedekind cut of I
i
in which (at least) one side of the cut
has conality .
Proof: Let I = (0 ) (1 ), I
i
= (0
i
) (1 ) ordered by (i, )
<
I
(j, ) i i < j or 0 = i = j and < or 1 = i = j and > .
Lemma 5.2: Suppose that T is categorical in > cf(), + [T[ < . If M K
,
then there exists a continuous increasing
F
- chain M
i
: i < cf()) of models such that
(1) M _
F
i<cf()
M
i
;
(2) |
i<cf()
M
i
| = ;
(3) +[T[ |M
i
| < |M
i+1
| <
(4) for each i < cf() , M
i
_
nice
j<cf()
M
j
.
Proof: If is a limit cardinal, choose a continuous increasing sequence
i
: i < cf()),
= sup
i<cf()
i
, +[T[
0
< . If is successor let
i
= 1 +i. Let I, I
i
: i < cf())) be
as in 5.1. By - categoricity of T WLOG M = EM(). Let M
i
= EM(I
i
) for i < cf().
Clearly (1) (2) and (3) hold. To obtain (4), observe that by 2.6 and 3.5 it suces to remark
that by demand (3) from 5.1 on I
i
: i < cf()) clauses () or () in 2.6 holds for each
t I I
i
.
Theorem 5.3. For every [+[T[, ] and M K
, there exists M
, M _
F
M
such that ()
M
: for every A [M
and N is nice.
Proof: The proof is by induction on .
Case 1: = + [T[. By 3.3 there is M
M _
F
M
and M
is nice. Given
A [M
[ let N = M
. Choose
a continuous increasing sequence
i
: i < cf()) such that if is a limit cardinal it is a
strictly increasing sequence with limit ; if is a successor, use
+
i
= and in both cases
+[T[
i
< . Find
M = M
i
: i < cf()) such that
35
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
(a) M _
F
i<cf()
M
i
(b) [[
i<cf()
M
i
[[ =
(c) [[M
i
[[ =
i
(d) M
i
_
nice
j<cf()
M
j
.
Why does
M exist? If = by 5.2, otherwise by 4.4 ( regular) and 4.11 ( singular).
Choose by induction on i < cf() models L
0
i
, L
1
i
, L
2
i
in that order such that
() M
i
_
F
L
0
i
_
F
L
1
i
_
F
L
2
i
K
i
;
() j < i L
2
j
_
F
L
0
i
;
() ()
L
1
i
holds , i.e. for each A [L
1
i
[, there is N K
+|T|+|A|
such that A
N _
F
L
1
i
and N is nice (so in particular L
1
i
is nice, letting A = [L
1
i
[);
() L
2
i
is nice and
i
- universal over L
1
i
;
() L
0
i
is increasing continuous;
() L
i
j<cf()
M
j
= M
i
(or use system of _
F
-embeddings).
For i = 0, let L
0
i
= M
0
. For i = j +1, note that by 2.1 there is an amalgam L
0
i
K
i
of M
i
, L
2
j
over M
j
since M
j
_
nice
M
i
and M
j
_
F
L
2
j
(use last phrase of 2.1 for clause ());
actually not really needed. For limit i, continuity necessitates choosing L
0
i
=
j<i
L
0
j
(note
that in this case L
0
i
=
j<i
L
2
j
). To choose L
1
i
apply the inductive hypothesis with respect
to
i
, L
0
i
to nd L
1
i
so that L
0
i
_
F
L
1
i
and ()()
(L
1
i
)
holds. To choose L
2
i
apply 3.10 to
L
1
i
K
i
giving L
1
i
_
F
L
2
i
, L
2
i
is nice and
i
-universal over L
1
i
(so () holds).
Let L =
i<cf()
L
0
i
=
i<cf()
L
1
i
=
i<cf()
L
2
i
, and let L
i
= L
0
i
if i is a limit, L
1
i
otherwise.
Now show by induction L
i
is nice [Why? show by induction on i for i = 0 or i successor
that L
i
= L
1
i
hence use clause (), if i is limit then L
i
is (
i)-saturated, hence L
i
is nice
by 4.8, 4.10]. Now L
i
: i < cf()) witnesses that if is regular, L is (, )-saturated by
4.4, if is singular, L is - saturated; in all cases L is - saturated of power , hence by
the results of section 4 (i.e. 4.8, 4.10) if < then L is nice.
Claim: M
= L is as required.
Proof: M _
F
i<cf()
M
i
_
F
i<cf()
L
0
i
= L K
i
, (AL
0
i
)N
2
j
N
0
i
_
F
L
0
i
(in particular N
2
j
_
F
N
0
i
); for limit i, N
0
i
=
j<i
N
0
j
. To choose N
1
i
, use ()
L
1
i
for the set A
i
= N
0
i
to nd a nice N
1
i
K
+|T|+|A|
, N
0
i
_
F
N
1
i
_
F
L
1
i
. Note that
|N
1
i
|
i
. Finally to choose N
2
i
note that by 3.9 the model N
1
i
has a nice extension
N
+
i
(of power |N
1
i
|) weakly universal over N
1
i
. Now N
1
i
is nice, hence N
2
i
is universal
over N
1
i
(by 3.6A(5)) and by 2.1 there is an amalgam N
i
of N
+
i
, L
1
i
over N
1
i
such that
|N
i
|
i
. Since L
2
i
is universal over L
1
i
one can nd an T - elementary submodel N
2
i
of
L
2
i
isomorphic to N
i
. Let N
i
be N
0
i
if i is a limit, N
1
i
otherwise; prove by induction on i
that N
i
is nice (by 4.2).
Now
i<cf()
N
0
i
is an T - elementary submodel of L of power at most + [T[ + [A[,
including A (by ()) and
i<cf()
N
0
i
is ( + [T[ + [A[, cf()) - saturated, hence (by 4.2)
nice, as required.
Corollary 5.4: Suppose that T is categorical in . Then every element of K
<
is nice.
Proof: Suppose otherwise and let N
0
K
<
be a model which is not nice. Choose a
suitable 0p such that |0p(N
0
)| and by 1.2 nd M
0
K
, N
0
_
F
M
0
_
F
0p(N
0
) i.e.
N
0
_
nice
M
0
. It follows that
(+) if N
0
_
F
N _
F
M
0
and N K
<
then N is not nice. [ Why? By 4.3; alternatively
suppose contrawise that N is nice. So there is N
1
K
<
, N
0
_
F
N
1
, N
0
nice
N
1
N
0
_
nice
N
since N
0
_
nice
M
0
and N _
F
M
0
, hence there is an amalgam N
K
<
of N
1
, N over N
0
. N
37
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
is nice, so N _
nice
N
; N
0
_
nice
N, so N
0
_
nice
N
and so N
0
_
nice
N
1
contradiction.] On the other
hand, applying 5.3 for = there exists M
satisfying ()
M
. By - categoricity
of T WLOG ()
M
0
holds (see 5.3) and for A = [N
0
[ yields a nice model N K
+|T|+N
0
such that N
0
_
F
N _
F
M
0
contradicting (+).
Corollary 5.5: Suppose that T is categorical in . Then /
<
has the amalgamation
property.
Proof: 2.1 and the previous corollary.
38
3
6
2
r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
2
-
0
5
m
o
d
i
f
i
e
d
:
1
9
9
8
-
1
2
-
0
5
References
[CK] Chang, C.C. and Keisler, H.J. Model theory. North-Holand (1973).
[D] Dickmann, M. Large innitary languages: model theory. North-Holland, (1975).
[D1] Dickman, M. Larger innitary languages, chapter IX of Model Theoretic Logics,
J. Barwise and S. Feferman editors, Perspectives in Mathematical Logic, Springer-
Verlag, New York Berlin Heidelberg Tokyo, 1985, 317364.
[HaSh323] Hart, B. and Shelah, S. Categoricity over P for rst-order T or categoricity for
L
can stop at
k
while holding for
0
, . . . ,
k1
, Israel J. Math. 67(1989) 117.
[HoSh109] Hodges, W. and Shelah, S. Innite reduced products, Annals of Math. Logic 20(1981),
77108.
[K] Keisler, H.J. Model theory for innitary logic. North-Holland, (1971).
[J] Jech, T. Set theory. Academic Press, (1978).
[L] Laver, R. On Fraisses order-type conjecture, Annals of Mathematics 93(1971), 89111.
[MaSh285] Makkai, M. and Shelah, S. Categoricity of theories in L
w
, with a compact cardinal,
Annals of pure and Applied Logic 47(1990), 4197.
[M] Morley, M. Categoricity in power, Trans. Amer. Math. Soc. 114(1965), 514518.
[N] Nadel, M. L