Saharon Shelah - Categoricity For Abstract Classes With Amalgamation

(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

CATEGORICITY FOR ABSTRACT
CLASSES WITH AMALGAMATION
SH394
Saharon Shelah
The Hebrew University of Jerusalem
Einstein Institute of Mathematics
Edmond J. Safra Campus, Givat Ram
Jerusalem 91904, Israel
Department of Mathematics
Hill Center-Busch Campus
Rutgers, The State University of New Jersey
110 Frelinghuysen Road
Piscataway, NJ 08854-8019 USA
Abstract. Let K be an abstract elementary class with amalgamation, and Lowen-
heim Skolem number LS(K). We prove that for a suitable Hanf number
0
if
0
<
0

1
, and K is categorical in
+
1
then it is categorical in
0
.
2000 Mathematics Subject Classication. 03C45, 03C75.
Key words and phrases. model theory, classication theory, nonelementary classes, categoric-
ity, Hanf numbers, Abstract elementary classes, amalgamation.
Partially supported by the United States-Israel Binational Science Foundation and I thank Alice
Leonhardt for the beautiful typing.
Publ. No. 394
Done 6,9/88
Revised after journal (in March 1999, in Oct. 2003-June 2004)
Latest Revision - 04/Oct/29
Typeset by A
M
S-T
E
X
1
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

2 SAHARON SHELAH
Annotated Content
I0 Introduction, pg.5-7
[We review background and some denitions and theorems on abstract ele-
mentary classes.]
I1 The Framework, pg.8-12
[We dene types, stability in , S(M) and E
: equivalence relations on
types all whose restrictions to models of cardinality are equal. We
recall that categoricity in implies stability in [LS(K), ).]
I2 Variant of Saturation, pg.13-16
[We dene <
,
and N is (, )-saturated over M and show universality
and uniqueness.]
I3 Splitting, pg.17-18
[We note that stability in implies that there are not so many -splittings.]
I4 Indiscernibility and E.M. models, pg.19-27
[We dene strong splitting and dividing, and connect them to the order
property and unstability.]
I5 Rank and Superstability, pg.28-33
[We dene one variant of superstability; in particular categoricity implies
it.]
I6 Existence of many non-splitting, pg.34-41
[We prove (e.g. for K categorical in = cf()) that if M
0
<
1
,
M
1

K
N
K
<
and p S(M
1
) does not -split over M
0
, then p can be extended to
q S(N) which does not -split over M
0
.
(Note: up to E
-equivalence the extension is unique). Secondly, if M

i
: i
) is
1
,
-increasing continuous in K
and p S(M
) then for some i we

have: p does not -split over M
i
.]
I7 More on Splitting, pg.42-44
[We connect non-splitting to rank and to dividing.]
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

CATEGORICITY, ETC. 3
II8 Existence of nice , pg.45-65
[We try to successively extend the ; of course, the we use which is proper
for linear orders such that we have as many denable automorphisms as
possible. We also relook at omitting types theorems over larger model (so
only restrictions will appear).]
II9 Small Pieces are Enough and Categoricity, pg.66-73
[The main claim is that for some not too large , if p
1
, p
2
S(M) are
E
-equivalent, |M| < where K is categorical in we have p

1
E
p
2

p
1
= p
2
.
Lastly, we derive that categoricity is downward closed for successor cardinals
large enough above LS(K).]
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

4 SAHARON SHELAH
0 Introduction
We try to nd something on
Cat
K
= : K categorical in
for K an abstract elementary class with amalgamation (see 0.1 below).
The Los conjecture = Morley theorem deals with the case where K is the class
of models of a countable rst order theory T. See [Sh:c] for more on rst order
theories. What for T a theory in an innitary language? (For a theory T, K is
the class K
T
= M : M [= T we may write Cat
T
instead of Cat
K
T
= Cat
K
).
Keisler gets what can be gotten from Morleys proof on L
1
,
0
. Then see
[Sh 48] on categoricity in
1
for L
1
,
0
and even L
1
,
0
(Q), and [Sh
87a], [Sh 87b] on the behaviour in the
n
s. Makkai Shelah [MaSh 285] proved:
if T L
,
0
, a compact cardinal then Cat
T

+
:
(2
+|T|
)
+ is empty
or is
+
:
(2
+|T|
)
+ (it relies on some developments from [Sh 300] but is
self-contained).
It was then reasonable to deal with weakening the requirement on to measura-
bility. Kolman Shelah [KlSh 362] proved that if Cat
T
where T L
,
(),
(2
)
+ where = [[ +, measurable, then (after cosmetic changes), for the right
T
the class M : M [= T, |M| < has amalgamation and joint embedding
property. This is continued in [Sh 472] which gets results on categoricity parallel
to the one in [MaSh 285] for the downward implication.
In [Sh 88] we deal with abstract elementary classes (they include models of
T L
,
0
, see 0.1), prove a representation theorem (see 0.5 below), and investigate
categoricity in
1
(and having models in
2
, limit models, realizing and materializ-
ing types). Unfortunately, we do not have anything interesting to say here on this
context. So we add amalgamation and the joint embedding properties thus getting
to the framework of Jonsson [J] (they are the ones needed to construct homoge-
neous universal models). So this context is more narrow than the ones discussed
above, but we do not use large cardinals. We concentrate here, for categoricity on
, on the case is successor >
(2
LS(K)
)
+. See for later works [Sh 576], [Sh 600],
[ShVi 635] and [Va02].
We quote the basics from [Sh 88] (or [Sh 576]).
We thank Andres Villaveces and Rami Grossberg and earlier Michael Makkai for
much help.
We thank John Baldwin for complaining repeatedly during 2003/2004 on 8,9 and
Alex Usvyatsov for help in proofreading their revisions, so 8.9 (and beginning of
9) were changed; elsewhere the changes are small.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

0.1 Denition. K = (K,
K
) is an abstract elementary class if for some vocabulary
= (K) = (K), K is a class of (K)-models, and the following axioms hold.
Ax0: The holding of M K, N
K
M depends on N, M only up to isomorphism
i.e. [M K, M

= N N K], and [if N
K
M and f is an isomorphism from
M onto the -model M
mapping N onto N
then N
K
M
].
AxI: If M
K
N then M N (i.e. M is a submodel of N).
AxII: M
0

K
M
1

K
M
2
implies M
0

K
M
2
and M
K
M for M K.
AxIII: If is a regular cardinal, M
i
(for i < ) is a
K
-increasing (i.e. i < j <
implies M
i

K
M
j
) and continuous (i.e. for limit ordinal < we have
M
=
_
i<
M
i
) then M
0

K
_
i<
M
i
K.
AxIV : If is a regular cardinal, M
i
(i < ) is
K
-increasing continuous and
M
i

K
N then
_
i<
M
i

K
N.
AxV : If M
0
M
1
and M

K
N for = 0, 1, then M
0

K
M
1
.
AxV I: LS(K) exists
1
; see below Denition 0.3.
0.2 Denition. 1) K
=: M K : |M| = .
2) We say h is a
K
-embedding of M into N is for some M
K
N, h is an isomor-
phism from M onto M
.
3) We say that K has amalgamation (or the amalgamation property) when if for
any models M
K for = 0, 1, 2 and
K
-embeddings h
of M
0
into M
for = 1, 2
there are M
3
, g
1
, g
2
such that M
3
K and g
is a
K
-embedding of M
into M
3
for
= 1, 2 and g
1
h
1
= g
2
h
2
.
4) K has the -amalgamation means that above M
for = 0, 1, 2 and = 3.
5) K has the point embedding property, JEP means that for any M
1
, M
2
K there
is M
3
K then
K
-embedding g
1
, g
2
of M
1
, M
2
into M respectively. The -joint
embedding property, JEP
means that above we assume M

1
, M
2
K
.
6) Let M <
K
N mean M
K
N & N ,= N.
0.3 Denition. 1) We say that is a Lowenheim Skolem number of K if

0
and:
()
K
for every M K, A M, [A[ there is M
, A M
K
M and
|M
| .
1
We normally assume M K M LS(K), here there is no loss in it. It is also natural to
assume |(K)| LS(K) which just means increasing LS(K).
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

6 SAHARON SHELAH
2) LS
(K) = Min : is a Skolem Lowenheim number of K.

3) LS(K) = LS
(K) +[(K)[.
0.4 Claim. 1) If I is a directed partial order, M
t
K for t I and
s <
I
t M
s

K
M
t
then
(a) M
s

K
_
tI
M
t
K for every s I
(b) if (t I)[M
t

K
N] then
_
tI
M
t

K
N.
2) If A M K, [A[ + LS
(K) |M|, then there is M

1

K
M such that
|M
1
| = and A M
1
.
3) If I is a directed partial order, M
t
N
t
K for t I and s
I
t M
s

K
M
t
& N
s

K
N
t
then
_
t
M
t

K
_
t
N
t
.
0.5 Claim. Let K be an abstract elementary class. Then there are
+
, such that:
(a)
+
is a vocabulary extending (K) of cardinality LS(K)
(b) is a set of quantier free types in
+
(each is an m-type for some m < )
(c) M K i for some
+
-model M
+
omitting every p we have
M = M
+

(d) M
K
N i there are
+
-models M
+
, N
+
omitting every p such that
M
+
N
+
, M = M
+
(K) and N = N
+
(K).
(e) if M
K
N and M
+
is an expansion of M to a
+
-model omitting every
p then we can nd a
+
-expansion of N omitting every p such that
M
+
N
+
.
0.6 Claim. Assume K has a member of cardinality
(2
LS(K)
)
+ (here and elsewhere
we can weaken this to: has a model of cardinality
for every < (2

LS(K)
)
+
).
Then there is proper for linear orders (see [Sh:c, Ch.VII,2]) such that:
(a) [()[ = LS(K)
(b) for linear orders I J we have
EM
(I, )
K
EM(J, )( K).
(c) EM
(I, ) has cardinality [I[ +LS(K) (so K has a model in every cardinality
LS(K)).
Proof. Here see 8.6.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

PART 1
1 The Framework
1.1 Hypothesis.
(a) K = (K,
K
) an abstract elementary class (0.1) so
K
= M K : |M| =
(b) K has amalgamation and the joint embedding property
(c) K has members of arbitrarily large cardinality, equivalently: K has a mem-
ber of cardinality at least
(2
LS(K)
)
+.
1.2 Convention. 1) So there is a monster C (see [Sh:a, Ch.I,1] = [Sh:c, Ch.I,1]).
1.3 Denition. 1) We say K (or K) is categorical in if it has one and only one
model of cardinality , up to isomorphism.
2) I(, K) is the number of models in K
(i.e., in K of cardinality ) up to isomor-

phism.
1.4 Denition. 1) We can dene tp( a, M, N) (when M
K
N and a N), as
( a, M, N)/E where E is the following equivalence relation: ( a
1
, M
1
, N
1
) E ( a
2
, M
2
, N
2
)
i M
K
N
, a

(N
) (for some but the same for = 1, 2) and M

1
= M
2
and there is N K satisfying M
1
= M
2
K
N and
K
-embedding f
: N
N
over M
(i.e. f M
is the identity) for = 1, 2 satisfying f

1
( a
1
) = f
2
( a
2
). We
can dene tp( a, , N) similarly.
2
2) We may omit N when N = C (see 1.2) and may then write
a
M
= a/M =
tp( a, M, C). We dene N is -saturated (when > LS(K)) by: if M
K
N, |M| < and p S
<
(M) (see below) then p is realized in M, i.e. for some
a N, p = tp( a, M, N).
3) S
(M) = tp( a, M, N) : a

N, M
K
N.
4) S(M) = S
1
(M) (we could have just as well used S
<
(M) =
_
n<
S
n
(M)).
5) If M
0

K
M
1
and p
(M
) for = 1, 2, then p
0
= p
1
M
0
means that for
some a, N we have M
1

K
N and a

N and p
= tp( a, M
, N) for = 1, 2. See
[Sh 300, Ch.II] or [Sh 576, 0] and see 1.10 below.
2
what about tp( a, A, N)? The cumbersomeness is that we end up dening essentially tp
( a
A, , N)
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

8 SAHARON SHELAH
1.5 Denition. Let K stable in mean: |M| [S(M)[ and LS(K).
1.6 Convention. If not said otherwise, is as in 0.6.
1.7 Claim. If K is categorical in and LS(K), then
(a) K is stable in every which satises LS(K) < , hence
(b) the model M K
is cf()-saturated (if cf() > LS(K)).

Remark. The rst proof below gives more and uses more.
Proof. Like [KlSh 362] but this is immersed with ultrapowers.
First Proof: So
3
let be as in 0.6. Now let I be such that:
(a) I is a linear order of cardinality
(b) for every J I, [J[ = there is J
1
satisfying
() J J
1
I
() [J
1
[ =
() if I
I is nite then for some automorphism g of I we have

g J = id
J
g(I
) J
1
.
(see [Sh 220, AP]).
Now suppose toward contradiction that M
0
K
, [S(M
0
)[ > , then we can
nd M
1
K
+ and a
i
M
1
for i <
+
such that M
K
M
1
and i < j <
+
tp(a
i
, M, M
1
) ,= tp(a
j
, M, M
1
). By 1.10(1) below we can nd M
2
K
such that M
1

K
M
2
. Let be as in 0.6. Now M
2
and EM
(K)
(I, ) are both
models in K
hence are isomorphic, so by renaming equal. So let a

i
=
i
( a
t
i
) with
t
i

n
i
I. By the pigeon-hole principle without loss of generality
i
=
, n
i
= n
and let J I, [J[ be such that M EM

(K)
(I, ) and let J
1
satisfy clauses
(b)()(), () above. For each i <
+
there is an automorphismf
i
of the linear order
I such that f
i
J = id
J
, f
i
(
t
i
) J
1
. So without loss of generality f
i
(
t
i
) = s. Now
each f
i
induces naturally an automorphism

f
i
of EM
u()
(I, ), which in particular
is an automorphism of EM
(K)
(I, ) = M
2
. This automorphism is the identity
3
the proof was written because of a requisition of Rami Grossberg!
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

on EM
()
(J, ) hence on M. Also

f
i
(a
i
) =

f
i
(
t
1
), . . . ,
(f
i
(
t
i
)) =
( s), so
is the same. Clearly tp(a
i
, M, M
1
) = tp(a
i
, M, M
2
) = tp(

f
i
(a
i
),

f
1
(M), M
2
) =
tp(
( s), M, M
2
), as this holds for every i <
+
we have gotten a contradiction.
Alternative proof: Let be as in 0.6. Assume N = EM
(, ) and M
K
N, M
K
. We can nd u , [u[ = such that M

K
EM
(u, ). Clearly it is enough

to show
() if

b
= (. . . , a
(,k)
, . . . )
k<k()
N for = 1, 2 as (, 0) < . . . <
(, k()) 1 and
( u)(k < k()))[((, k) < (2, k) < )
((1, k)) > (2, k) > )]
then tp(b
1
, M, N) = tp(b
2
, M, N).
But () is immediate: let N
i
= EM
(u
, ) where u
i
= u (, k) : k < k()
for = 1, 2, so there is an isomorphism f from EM(u
1
, ) onto EM(u
2
, ) which is
the identity on a
: u and maps a
(1,k)
to a
(2,k)
for k < k(). So f can be
extended to an automorphism f
+
of C, f
+
M = id
, f
+
(b
1
) = b
2
.
So if M K
, [S(M)[ > , there is M

+
such that M
K
M
+
K
+ and
[tp(b, M, M
+
) : b M
+
[ =
+
. Let N
+
K
be such that M
+
K
N
+
K
.
So there is an isomorphism g from N
+
onto N. Now g(M), g(M
+
) contradicts
what we have proved above.
Similarly LS(K) = < , M K
[S
(M)[ . So we have proved

clause (a).
Now for proving clause (b); it just follows from clause (a).
1.7
1.8 Denition. 1) For LS(K), E
= E
1
[K], E
is the following relation,

p E
q i for some M K, m < we have

p, q S
m
(M) and [N
K
M & |N| p N = q N].
2) We say p S
m
(M) is -local if p/E
is a singleton.
3) We say K is -local if every p S
<
(M) is -local.
4) We say c realizes p/E
in M
if M
K
M
, c M
and [N
K
M & |N|
tp(c, N, M
) = p N].
1.9 Remark. 0) Obviously E
is an equivalence relation.
1) In previous contexts E
LS(K)
is equality, e.g. the axioms of NF in
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

10 SAHARON SHELAH
[Sh 300, Ch.II,1] implies it; but here we do not know this is the main diculty.
We may look at this as our bad luck, or inversely, a place to encounter some of the
diculty of dealing with L
,
(in which our context is included).
2) Note that the -local does not imply -compactness which means: if m <
, M K and p = p
N
: N
K
M, |N| ), p
N
S
m
(N) and [N
1

K
N
2

K
M & |N
2
| p
N
1
= p
N
2
N
1
] then there is p S
m
(M) such that
N
K
M & |N| p N = p
N
.
1.10 Claim. 1) There is no maximal member in K, in fact for every M K there
is N, M <
K
N K, |N| |M| + LS(K), hence for every |M| + LS(K) there
is N K
such that M <

K
N K
.
2) If p
2
S
(M
2
) and M
1

K
M
2
K then for one and only one p
1
S
(M
1
)
we have p
1
= p
2
M
1
.
3) If p
1
S
(M
1
) and M
1

K
M
2
K then for some p
2
S
(M
2
) we have
p
1
= p
2
M
1
.
4) If M
1

K
M
2

K
M
3
and p
(M
) for = 1, 2, 3 then p
3
M
2
= p
2
&
p
2
M
1
= p
1
p
3
M
1
= p
1
.
Proof. 1) Immediate by clause (c) of the hypothesis 1.1 and claim 0.6.
2) Straightforward.
3) By amalgamation.
4) Check.
1.10
1.11 Claim. If M
i
: i ) is
K
-increasing continuous and p
n
S
(M
n
) and
p
n
= p
n+1
M
n
for n < , then there is p
(M
) such that n < p

M
n
= p
n
.
Proof. Let N
0
be such that M
0

K
N
0
and p
0
= tp(a, M
0
, N
0
). We now choose
(N
n
, h
n
) by induction on n such that:
(a) N
n
K is
K
-increasing
(b) h
n
is a
K
-embedding of M
n
into M
n
(c) h
n
increases with n
(d) tp(a, h
n
(M
n
), N
n
) is h
n
(p
n
).
For n = 0, (N
n
, id
M
0
) are as required.
For n + 1, use p
n+1
M
n
= p
n
and straight chasing diagrams.
1.11
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

1.12 Remark. In 1.11 we do not claim uniqueness and do not claim existence
replacing for of uncountable conality. In general not true [Saharon add].
Compare with 8.5, 9.2.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

12 SAHARON SHELAH
2 Variant of Saturated
2.1 Denition. Assuming K stable in and is an ordinal <
+
, means
ordinal product.
1) M <
,
N if: M K
, N K
, M
K
N and there is a
K
-increasing sequence
M = M
i
: i ) which is continuous, M
0
= M, M

K
N and every
p S
1
(M
i
) is realized in M
i+1
.
2) We say M <
1
,
N i M K
, N K
, M
K
N and there is a
K
-increasing
sequence

M = M
i
: i ), M
0
= M, M
= N and every p S
1
(M
i
) is
realized in M
i+1
.
3) If = 1, we may omit it.
2.2 Lemma. Assume K stable in and <
+
.
0) If 0, 1 and
1
<
2
<
+
and there is b
2
such that otp(b) =
1
and [ = 1 b unbounded in
2
] then <
,
2
<
,
1
.
1) If M K
, then for some N we have M <
,
N and for some N, M <
1
,
N.
2) (a) If M K
, M
K
M
,
N then M
,
N.
(b) If M K
, M
K
M
,
N
K
N K
then M
,
N (so
1
,
,
).
3) If M
i
: i < ) is
K
-increasing sequence in K
, M
i

M
i+1
and <
+
is a
limit ordinal, then M
0

1
,
_
i<
M
i
.
4) If M
N then:
(a) any M
can be
K
-embedded into N (here we can weaken |M
| =
to |M
| )
(b) If M
K
N
, h is a
K
-embedding of M
into M then h can be

extended to a
K
-embedding of N
into N.
5) If M
1
,
N
for = 1, 2, h an isomorphism from M

1
into [onto] M
2
then h
can be extended to an isomorphism from N
1
into [onto] N
2
.
6) If M
1
,
N
for = 1, 2 then N
1

= N
2
(even over M).
7) If M
,
N, M
K
M
then M
can be
K
-embedded into N over M.
8) If > LS(K) and M <
1
,
N then N is cf()-saturated.
Proof. See [Sh 300, Ch.II,3.10,p.319] and around, we shall explain and prove part
(8) below.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

2.3 Discussion: There (in [Sh 300, Ch.II,3.6]) the main point was that for >
LS(K), the notions -homogeneous universal and -saturation (i.e., every small
1-type is realized) are equivalent.
Not hard, still [Sh 300, Ch.II,3.6] was a surprise to some (including myself). In
rst order the equivalence saturated homogeneous universal for seemed, with
a posteriori wisdom, natural as the homogeneity used was anyhow for sequences of
elements realizing the same rst order formulas so (forgetting about the models)
to some extent this seemed natural; i.e. asking this for any type of 1-element was
very natural.
But here, types of 1-element are really meaningful only over a model. So it seems
that if over any small submodel every type of 1-element is realized (say in A) and
we would like to embed N
K
N
0
, N
0

K
A into A over N
0
, we encounter the
following problem: we cannot continue this as after stages, as we get a set which
is not a model (if LS(K) >
0
this absolutely necessarily fails; and if LS(K) =
0
at
best the situation is as in [Sh 87a]).
This explains a natural preconception making you not believe; i.e. psychological
barrier to prove. It does not mean that the proof is hard.
Note that in [Sh 48], [Sh 87a], [Sh 87b] and even [Sh 88] the types are a still set of
formulas and essentially (after cleaning) rst order.
2.4 Remark. Note that
1
,
, regular are the interesting ones as
1
,
=
1
,cf()
.
[Why? For limit ordinal <
+
,
1
,
1
,cf()
by 2.2(0) and equality holds by the
uniqueness 2.2(6).]
Still
0
,
is enough for universality (2.2(4)) and is natural,
1
,
is natural for
uniqueness. BUT <
1
,
0
=<
1
,
1
can be proved only under categoricity (or something
like superstability assumptions). For understanding this we may consider a rst
order T stable in . Then, M <
1
,
N is equivalent to:
(i) |M| = |N| = , M, N [= T
(ii) and there is M
i
: i ) which is -increasing continuous such that
() M
0
= M M
= N
() (M
i+1
, c)
cM
i
is saturated.
Question: Now, is N saturated when M <
1
,
N?
Answer: It is saturated i cf()
r
(T). See [Sh:c, Ch.III,3].
This is similar to S-limit models for S a stationary subset of
+
, see [Sh 88].
That is, if M <
1
,
N then N is a <
+
: cf() = -limit model. If T is
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

14 SAHARON SHELAH
rst order stable in , then there are such models for every ; for = 1
we get the saturated model. Of course, N is a superlimit model (see [Sh 88]) if
= cf() (M)(M
1
,
N).
Before we prove 2.2(8), recall
2.5 Denition. M K is -saturated if > LS(K) and:
N
K
M, |N| < , p S
1
(N) p realized in M.
Proof of 2.2(8).
Statement: If M <
1
,
N ( regular) then N is -saturated.
Note: if LS(K) the conclusion is essentially empty, but there is no need for the
assumption > LS(K).
Proof. Let

M = M
i
: i ) witness M
1
,
N so M
0
= M, M
= N, M
i

K
-
increasing continuous and every p S(M
i
) is realized in M
i+1
.
Assume
() N
K
N, |N
| < , p S(N
).
We should prove that p is realized in N. But M
i
: i ) is increasing
continuous
cf( ) = > |N
|
so N
K
M
=
_
i<
M
i
implies there is i() < , such that N
M
i()
hence by Axiom V we have N
K
M
i()
. So p has (by amalgamation, i.e., 1.10(3))
an extension p
S(M
i()
), i.e., p
N = p. By the choice of M
i
: i ), p
is realized in M
i()+1
so in M
= N and the same element realizes p by the

denition of we are done.
2.2
Comment: Hence length (instead of ) suces.
But for the uniqueness seemingly
4
it does not. See 2.2(4) + (5).
Comment: The denitions of
0
,
,
1
,
are also essentially taken from
[Sh 300, Ch.II,3.10]. We need the intermediate steps to construct models so we
have to have of them in order to deal with all the elements.
4
but see [Sh 600, 4] it suces
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

2.6 Claim. If K is categorical in , M K
and cf() > then:

if N <
K
M K
, N K
, N
<
K
M, h an isomorphism from N onto N
, then h
can be extended to an automorphism of M.
Proof. By 1.4 we have LS(K) < K stable in . We can nd M
i
: i < )
which is <
K
-increasing continuous, |M
i
| = [i[ + LS(K) and M
i
<
1
|i|+LS(K),|i|+LS(K)
M
i+1
. By the categoricity assumption without loss of generality M =
_
i<
M
i
. As
cf() > for some i
0
< we have N, N
M
i
0
.
By 2.2(5) we can build an automorphism.
2.6
To restate in later names
2.7 Denition. For LS(K), we say N K
is (, )-brimmed if for some M

we have M <
1
,
N (so is , normally regular); we then say N is (, )-brimmed
over M.
Restating the earlier statements
2.8 Claim. 1) The (, )-brimmed model is unique (even over M) if it exists at
all.
2) If M is (, )-brimmed, = cf() > LS(K) then M is -saturated.
3) If M is (, )-brimmed for every = cf() and > LS(K) then M is
-saturated.
2.9 Discussion: It is natural to dene saturated as |M|-saturated. (It may cause
confusions using the closely related notion of being (, )-brimmed for every regular
.) This is particularly reasonable when the cardinal is regular, e.g. if K
categorical in , = cf() the model in K
is -saturated.
Part of the program is to prove that all the denitions are equivalent in the
superstable case.
For now in Denition 2.7 we have not said when such a model exists; stability
in (for our classes which has amalgamation and JEP) is sucient and necessary.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

16 SAHARON SHELAH
3 Splitting
Whereas non-forking is very nice in [Sh:c], in more general contexts, non rst order,
it is not clear whether we have so good a notion, hence we go back to earlier notions
from [Sh 3], like splitting. It still gives for many cases p S(M), a denition of
p over some small N
K
M. We need -splitting because E
LS(K)
is not known to
be equality (see 1.8). We concentrate (in Denition 3.2 below) on the case N
1
, N
2
are models not sequences as in this work this is the most useful case (though those
sequences can be of length < |N|)
3.1 Context. Inside the monster model C.
3.2 Denition. p S(M) does -split over N
K
M if:
|N| , and there are N
1
, N
2
, h such that:
|N
1
| = |N
2
| and N
K
N

K
M, for = 1, 2
h an elementary mapping from N
1
onto N
2
over N such that
the types p N
2
and h(p N
1
) are contradictory (equivalently distinct).
3.3 Claim. 1) Assume K is stable in , LS(K). If M K
and p S
1
(M),
then for some N
0
M, |N
0
| = , p does not -split over N
0
(see Denition 3.2).
2) Moreover, if 2
> , M
i
: i + 1) is <
K
-increasing, a
m
(M
+1
),
tp( a, M
i+1
, M
+1
) does ( )-split over M
i
, then K is not stable in .
Proof of 3.3. 1) If not, we can choose by induction on i < N
i
, N
1
i
, N
2
i
, h
i
such
that:
(a) N
i
: i ) is increasing continuous, N
i
<
K
M, |N
i
| =
(b) N
i

K
N
i

K
N
i+1
(c) h
i
is an elementary mapping from N
1
i
onto N
2
i
over N
i
,
(d) p N
2
i
, h
i
(p N
1
i
) are contradictory, equivalently distinct (we could have
dened them for i <
+
).
Let = Min : 2
> so 2
<
. Now contradict stability in as in part (2).
2) Similar to [Sh:a, Ch,I,2] or [Sh:c, Ch.I,2] (by using models), but we give details.
Without loss of generality M
i
K
for i + 1. For each i < let N

i,1
, N
i,2
be such that M
i

K
N
i,

K
M
i+1
, g
i
an isomorphism from N
i,1
onto N
i,2
over M
i
and tp( a, N
i,2
, M
i+1
) ,= g
i
(tp( a, N
i,1
, M
i+1
). Without loss of generality 2
<
.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

We dene by induction on a model M
and for each

2, a mapping h
such that:
(a) M
is
K
-increasing continuous
(b) for

2, h
is a
K
-embedding of M
into M
(c) if < ,

2, then h
(d) if = + 1,

2, then h
<0>
N
i,1
= h
<1>
N
i,2
.
There is no problem to carry the denition (we are using amalgamation only in
K
and if we start with M

0
K
only in K
). Now for each

2 we can
nd M
, M

K
M
and
K
-embedding h
+
of M
+1
into M
extending
h
=
_
<
h
. Now tp(h
+
( a), M
, M
) :

2 is a family of 2
> distinct
members of S
m
(M
) and recall M
so we are done.
3.3
3.4 Conclusion. [Assume the conclusion of 3.2]. If p S
m
(M), M is
+
-saturated,
= cf() , then for some N
0
<
,
N
1

K
M, (so |N
1
| = ) we have:
p is the E
-unique extension of p N
1
which does not -split over N
0
, which means:
if q S(M), q N
1
= p N
1
and p does not -split over N
0
, then pE
q.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

18 SAHARON SHELAH
4 Indiscernibles and E.M. Models
4.1 Notation. We can below replace h
i
by the sequence h
i
(t) : t Y ).
4.2 Denition. Let h
i
: Y C for i < i
.
1) h
i
: i < i
) is an indiscernible sequence (of character < ) (over A) if for every g,

a partial one to one order preserving map from i
to i
(with domain of cardinality

< ) there is f AUT(C), such that
g(i) = j h
j
h
1
i
f
(and id
A
f).
So omitting means > i
.
2) h
i
: i < i
) is an indiscernible set (of character < ) (over A) if: for every g, a

partial one to one map from i
to i
(with [Dom(g)[ < ) there is f AUT(C),

such that
g(i) = j h
j
h
1
i
f
(and id
A
f).
3) h
i
: i < i
) is a strictly indiscernible sequence, if i
and for some , proper

for linear orders (see [Sh:a, Ch.VII] or [Sh:c, Ch.VII]) in vocabulary
1
= ()
extending (K), there are M
1
= EM
1
(i
, ) with skeleton x
i
: i < i
) (so M
1
is
the Skolem Hull of x
i
: i < i
which is an indiscernible sequence for quantier

free formulas), and there is a sequence of unary terms
t
: t Y ) such that:
t
(x
i
) = h
i
(t) for i < i
, t Y
M
1
(K) <
K
C.
4) Let h
i
: Y
i
C for i < i
we say that h
i
: i < i
) has localness if ( is a
cardinal and):
() if h
i
: Y
i
C for i < i
and for every u [i
]
<
there is an automorphism
f
u
of C such that f
u
A = id
A
and i u f
u
h
i
= h
i
, then there is an
automorphism f of C such that f A = id
A
and i < i
f h
i
= h
i
.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

4.3 Denition. 1) K has the (, )-order property if for every there are A C
and a
i
: i < ), where a
i

C and [A[ such that:
() if i
0
< j
0
< , i
1
< j
1
< then for no f AUT(C) do we have
f A = id
A
, f( a
i
0
a
j
0
) = a
j
1
a
i
1
.
If A = i.e. = 0, we write -order property.
2) K has the (
1
,
2
, ) order property if for every there are A C satisfying
[A[ , a
i
: i < ) where a
i

1
C and
b
i
: i < ) where

b
i

2
C such that
() if i
0
< j
0
< , i
1
< j
1
< , then for no f AUT(C) do we have
f A = id
A
, f( a
i
0
) = a
j
1
, f(
b
j
0
) =

b
i
1
.
4.4 Observation. So we have obvious monotonicity properties and if we can
let A = ; so the (, )-order property implies the ( +)-order property.
4.5 Claim. 1) Any strictly indiscernible sequence (over A) is an indiscernible se-
quence (over A).
2) Any indiscernible set (over A) is an indiscernible sequence (over A); can add
of character < .
Proof. Obvious.
4.6 Claim. 1) If LS(K) + [Y [ and for each <
(2
)
+ we have h
i
: Y C,
for i < (e.g. h
i
= h
i
) then for any innite i
, we can nd h
j
: j < i
), a strictly
indiscernible sequence, with h
j
: Y C such that:
() for every n < , j
1
< < j
n
< i
for arbitrarily large <

(2
)
+ we can
nd i
1
< < i
n
< and f AUT(C) such that h
(h
)
1
f.
2) If in part (1) for each , the sequence h
j
: j < ) is an indiscernible sequence
of character
0
, in () any i
1
< < i
n
< i
will do.
3) In Denition 4.3 we can restrict to <
(2
++LS(K)
)
+ and get an equivalent
version.
4) In Denition 4.3(1) we can demand a a
i
: i < ) is strictly indiscernible (where
a lists A) and get an equivalent version. Similarly in 4.7(2).
5) If LS(K) + [Y [, N
K
C and for each <
(2
)
+ we have h
i
: Y N for
i < and N
1
is an expansion of N with [(N
1
)[ , then for some expansion N
2
of N
1
with [(N
2
)[ and we have:
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

20 SAHARON SHELAH
(a) () = (N
2
)
(b) for linear orders I J we have
EM
(K)
(I, )
K
EM
(K)
(J, ) K
and the skeleton of EM
(K)
(I, ) is a
t
: t I), a
t
= a
t,y
: y Y )
(c) for every n < for arbitrarily large <
(2
)
+ for some i
0
< . . . i
n1
< ,
for every linear order I and t
0
< < t
n1
in I, letting J = t
0
, . . . , t
n1
there is an isomorphism g from EM(J, ) EM(I, ) (those are (N

2
)-
models) onto the submodel of N
2
generated by
_
<n
Rang(h
) such that
h
(y) = g(a
t,y
).
Proof. As in [Sh:c, Ch.VII,5] and [Sh 88], see 8.7 for a similar somewhat more
complicated proof.
4.6
As in the rst order case:
4.7 Lemma. 1) If there is a strictly indiscernible sequence which is not an in-
discernible set of character
0
called a
i
: i < ), then K has the [g(a
i
)[-order
property.
Remark. Permutation of innite sets is a more complicated issue. That is, assume
a
i
: i < i
) is a strictly indiscernible sequence over A of character

+
but is
not an indiscernible set over A of character
+
and i

+
. Does K have the
(g( a
0
), [A[ + g( a
0
))-order property.
4.8 Claim. 1) If K has the -order property then:
I(, K) = 2
for every > ( + LS(K))

+
(and other strong non-structure properties).
2) If K has the (
1
,
2
, )-order property and =
1
+
2
+ then for some
M K
, we have [S
2
(M)/E
[ > .
Proof. 1) By [Sh:e, Ch.III,3] (preliminary version appears in [Sh 300, Ch.III,3])
(note the version on e.g., (L
+
,
)).
2) Straight.
4.8
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

4.9 Denition. 1) Suppose M
K
N and p S
m
(N). Then p divides over M if
there are elementary maps h
i
: i < ), Dom(h
i
) = N, h
i
M = id
M
, h
i
: i < )
is a strictly indiscernible sequence and h
i
(p) : i < is contradictory i.e. no
element (in some C
, C <
K
C
) realizing all of them; recall is the cardinality of C.

Let -divides mean no elements realize of them.
2)
(K) [or
(K)] is the set of regular such that for some

K
-increasing contin-
uous M
i
: i +1) in K
and b M
+1
for every i < we have: tp(b, M
, M
+1
)
[or tp(b, M
i+1
, M
+1
)] divides over M
i
; so .
3)
,
(K) [or
,
(K)] is the set of regular such that for some
K
-increasing
continuous sequence M
i
: i + 1) in K
and b M
+1
for every i < we have:
tp(b, M
, M
+1
) [or tp(b, M
i+1
, M
+1
)], -divides over M
i
, so (see Denition
4.12 below).
4.10 Remark. 1) A natural question: is there a parallel to forking?
2) Note the dierence between
(K) and
(K), e.g., 4.11(2) is not clear for
(K).
Note that now the local character is apparently lost.
4.11 Fact. 1) In Denition 4.9(1) we can equivalently demand: no element realizing

(2
)
+ of them, where = |N|.
2) If
(K), = cf() then
(K) and similarly of
,
(K).
3)
(K)
(K) similarly
,
(K)
,
(K).
4.12 Denition. Suppose M
K
N, p S(N), M K
, LS(K).
1) We say p does -strongly split over M, if there are a
i
: i < ) such that:
(i) a
i

C for i < , <
+
, a
i
: i < ) is strictly indiscernible over M
(ii) for no b realizing p do we have tp( a
0
b), M, C) = tp( a
1
b), M, C).
2) We say p explicitly -strongly splits over M if in addition a
0
a
1
N.
3) Omitting means any (equivalently = |N|).
4.13 Claim. 1) Strongly splitting implies dividing with models of cardinality
if ()
holds where ()
= ()
,
0
,
0
and
()
,,
If a
i
: i < i
) is a strictly indiscernible sequence, a

i

C,
b
>
C, then for
some u i
, [u[ < and the isomorphism type of (C, a

i
b) for all i i
u
is the same.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

22 SAHARON SHELAH
4.14 Claim. 1) Let () = + + LS(K). Assume a
i
: i < i
) and

b form
a counterexample to ()
,,
of 4.13 and
(2
()
)
+ then K has the ()-order
property.
2) We can also conclude that for every + LS(K), for some M K
we have
[S
g(
b)
(M)[ > , note g(
b) < .
3) If in (1) we have <
(2
()
)
+ we can still get that for every ++LS(K)+
for some M K
, we have [S
g(
b)
(M)[
, moreover [S
g(
b)
(M)/E
.
4) In part (1) it suces to have such an example for every <
(2
()
)
+, of course,
for a xed ().
Proof. Straight, using 4.15 below.
4.15 Claim. Assume M = EM
(K)
(I, ),LS(K) +g( a
t
) for t I, [[ and
M
K
N,
b

N and
() for no J I, [J[ <
(2
)
+ do we have for all t, s IJ,
tp( a
t
b, , N) = tp( a
s
b, , N).
Then
(A) we can nd
proper for linear orders and a formula (not necessarily

rst order, but is preserved by
K
-embeddings) such that for any linear
order I
letting M
= EM(I
) having the skeleton a
t
: t I), a
t
= a
t
b
t
, g( a
t
)
, g(
b
t
) = and we have:
M
[= [ a
t
,
b
s
] t < s
(if < , this is half the nitary order property)
(B) this implies instability in every
if <
(C) this implies the -order property and even the (, [[, 0)-order property
(D) if we strengthen the assumption to

b

M then [J[ <
+
and just [J[ <
[[
+
+
0
in () suces
(E) if , for some N
, then [S
(N
)[ > moreover [S
(N
)/E
[ >
.
Proof. As we can increase I, without loss of generality the linear order I is dense
with no rst or last element and is (
(2
)
+)
+
-strongly saturated, see Denition
4.18 below. So for some p and some interval I
0
of I, the set Y
0
= t I
0
:
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

tp( a
t
b, , N) = p is a dense
5
subset of I
0
. Also for some q S
(M)p, the
set Y
1
= t I : tp( a
t
b, , N) = q has cardinality
(2
)
+ and let Y
1
Y
1
have
cardinality
(2
)
+. As we can shrink I
0
without loss of generality I
0
is disjoint from
Y
1
and as we can shrink Y
1
without loss of generality (s Y
1
)(t I
0
)(s <
I
t) or
(s Y
1
)(t I
0
)(t <
I
s).
By the Erdos-Rado theorem, for every <
(2
)
+ there are s
1
for <
such that s
: < ) is strictly increasing or strictly decreasing; without loss of

generality the case does not depend on , so as we can invert I without loss of
generality it is increasing. Hence (try (p
1
, p
2
) = (p, q) and (p
1
, p
2
) = (q, p), one will
work)
() we can nd p
1
,= p
2
such that
() for every <
(2
)
+ there is an increasing sequence t
: < + ) of
members of I such that
(i) < tp( a
t
b, , N) = p
0
(ii) < + tp( a
t
b, , N) = p
1
.
[Note that we could have replaced increasing by
(iii) < < t
<
I
t
<
I
t
+
<
I
t
+
.
Why? Let I
1
= t I : ( < ) s
< t, so every A I
1
of cardinality
(2
)
+
has a bound from below in I
1
, so for some q
1
S
(M) the set I

2
= t I
1
:
tp( a
t
b, , N) = q
1
is unbounded from below in I
1
. If q
1
,= q then q
1
, q can serve
as p
1
, p
2
, so assume q
1
= q, so p, q
1
can serve as p
1
, p
2
.]
For every <
(2
)
+ for every < we can nd an automorphism f
,
of C such
that f
,
( a
t
) is a
t
if < and is a
t
+
if [, ), and let

b
= f
1
,
(
b). So in
C, for <
(2
)
+, we have ( a
t
) : < ) satises tp( a

t
, , C) is p
1
i it is
,= p
0
i .
Now we apply 4.6(5) with h
i
listing a
and letting N
1
be EM(I, ) (so (N
1
) =
()) and we get as there.
So we have proved clause (A) and clause (B) by 4.8(2), by easy manipulations
we get clause (E) and so (C).
We are left with clause (D). Clearly there is

t = t
i
: i < i
) satisfying i
< [[
+
+
0
such that

b = b
: < ), b
( a
t
i(,0)
, . . . , a
t
i(,n()1)
) where i(, ) < i
5
as {tp( a, M, C) : a

C} is 2
M+||+LS(K)
by the amalgamation property and the choice
of C.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

24 SAHARON SHELAH
a ()-term. So by the version of () used in clause (D), necessarily for some
s
1
, s
2
IJ we have:
p
1
,= p
2
where
p
1
= tp( a
s
1
b, , N)
p
2
= tp( a
s
2
b, , N)
Clearly s
1
,= s
2
. By renaming without loss of generality s
1
<
I
s
2
and initial seg-
ments J
( 3) of J we have = J
0
J
1
J
2
J
3
= J and for every t J, t <
I
s
1
t J
1
and t <
I
s
2
t J
2
. So for some 0 = i
0
i
1
i
2
i
3
= i
we have
t
i
<
I
s
1
i < i
1
and s
1
<
I
t
i
<
I
s
2
i
1
i < i
2
and s
2
<
I
t
i
i
2
< i < i
3
.
As I is (
(2
)
+)
+
-strongly saturated we can increase J so adding to J (by the sat-
uration of I) without loss of generality < & < n() i(, ) / t
1
, t
2
, and
J
= t J : t t
for = 1, 2 so t
< s
for = 1, 2. So for every linear order I
we can dene a linear order I
with set of elements

J
1
(JJ
2
) (s, t) : s I
, t J
2
J
1
linearly ordered by:

(a) on J
1
(JJ
2
) as in J
(b) t
1
< (s
, t
) < (s
, t
) < t
2
if t
1
J
1
, t
2
J
3
J
2
, s
, s
and (t
, t
J
2
J
1
and (s
<
I
) (s
= s
& t
<
J
t
).
In M = EM(I
, ) dene, for s I
c
s
= a
s,t
b
s
=
( c
s,i(,0)
, c
s,i(,1)
, . . . , c
s,i(,n()1)
) : < ).
Easily
s
<
I
tp( a
s

b
s
, , M) = p
1
s
tp( a
s

b
s
, , M) = p
2
.
By easy manipulations we can nish.
4.15
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

4.16 Claim. Assume K is categorical in and
(a) 1 and LS(K) < = cf() and
( < )([[
< )
(b) a
i

C for i < .
Then for some W of cardinality , the sequence a
i
: i W) is strictly indis-
cernible.
Proof of 4.16. Let M
C, |M
| = and < a
. There is
M
, M
C, |M
| = . So M

= EM(, ) and without loss of gen-
erality equality holds. So there is u , [u[ such that M
EM(u, ).
Hence without loss of generality M
= EM(u, ). So a

EM(u
, ) for some
u
u, [u
[ .
Without loss of generality: otp(u
) = j
, so for < , OP
u
,u
, the order preserv-

ing map from u
onto u
, induces f
,
: EM(u
, )
iso
onto
EM(u
, ), and without
loss of generality f
,
( a
) = a
.
Now as u is well ordered and the assumption (a), (or see below) for some w []
the sequence u
: w) is indiscernible in the linear order sense (make them a

sequence). Now we can create the right .
[Why? Let u
=
,j
: j < j
where
,j
increases with j. For < , let
A
=
,j
: < , j < j

_
<,j
,j
+ 1. Let
,j
= Min A
:
,j

and for each S
0
= < : cf() > let h() = Min < :
,j
A
(note
that A
: ) is increasing continuous, cf() > [j
[ and
,j
A
by the
denition of the A
s).
By Fodors lemma for some stationary S
1
S
0
, h S
1
is constantly
. As
( < )([[
< = cf()) for some S

2
S
1
for each j < j
and for all S

2
,
the truth value of
,j
A
(e.g.
,j
=
,j
) is the same and
,j
: S
2
) is
constant. Now u
: S
2
) is as required.]
4.16
That is, see more [Sh 620, 7].
4.17 Observation. If = cf() and ( < )([[
< ) and j
< and
,j
is an
ordinal for < , j < j
then for some stationary set S < : cf() =

+
the
sequence
,j
: j < j
) : S) is indiscernible.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

26 SAHARON SHELAH
4.18 Denition. A model M is -strongly saturated if:
(a) M is -saturated
(b) M is strongly -homogeneous which means: if f is a partial elementary
mapping from M to M, [Dom(f)[ <
then (g AUT(M))(f g).
4.19 Remark. 1) If =
<
, I a linear order of cardinality , then there is a
-strongly saturated dense linear order J, I J.
2) We can even get a uniform bound on [J[ (which only depends on ).
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

5 Rank and Superstability
5.1 Denition. For M K
, p S
m
(M) (and LS(K), of course) we dene
R(p), an ordinal or as follows: R(p) i for every < there are M
+
, M
K
M
+
K
, p p
+
S
m
(M
+
), R(p
+
) & [p
+
-strongly splits over M]. In
case of doubt we write R
. This is well dened and has the obvious properties:

(a) monotonicity, i.e., p
1
= p
2
M
1
R(p
1
) R(p
2
)
(b) if M K
, p S
m
(M) and Rk(p) then for some N, q satisfying
M
K
N K
and q S
m
(N) we have: q M = p and Rk(q) =
(c) automorphisms of C preserve everything
(d) the set of values is [0, ) or [0, ) for some < (2
)
+
, etc.
5.2 Denition. We say K is (, 1)-superstable if:
M K
& p S(M) R(p) <

_
equivalently < (2
)
+
_
.
5.3 Claim. If ()
from 4.13 above fails, then (, 1)-superstability fails.

Proof. Straight.
5.4 Claim. If K is not (, 1)-superstable, then there are a sequence
M
i
: i + 1) which is <
K
-increasing continuous in K
and m < and

a
m
(M
+1
) such that (i < )
_
a
M
i+1
does -strongly split over M
i
.
Also the inverse holds.
Proof. As usual.
5.5 Claim. 1) If K is not (, 1)-superstable then K is unstable in every such
that
0
> + + 2
0
.
2) If
(K) and
> LS(K), then K is not -stable.

Remark. We intend to deal with the following elsewhere; we need stable amalga-
mation
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

28 SAHARON SHELAH
(A) if
(K) and
> =
LS(K) or just there is a tree with nodes

and > -branches and LS(K), then K is not stable even modulo E
(B) if
(K), cf() = and
<
,
then K is not -stable.
5.6 Remark. 1) Clearly 5.5(1) this implies I(LS(K)
+((
0
+)+n)
, K) [[ when
=
0
. We conjecture that [GrSh 238] can be generalized to the context of (1)
with cardinals which exists by ZFC.
2) Note that for complete rst order stable theory T, K = MOD(T) so
K
=
MOD(T), for regular we have ()
1
()
2
where
()
1
for any increasing chain M
i
: i < ) of -saturated models of length
, the union
_
i<
M
i
is -saturated,
()
2

r
(K).
From this point of view, rst order theory T is a degenerated case:
r
(T) is an
initial segment so naturally we write the rst regular not in it. This is a point
where [Sh 300] opens our eyes.
3) In fact in 5.5 not only do we get |M| = , [S(M)[ > but also [S(M)/E
[ > .
4) Let me try to explain the proof of 5.5, of course, being inuenced by the rst
order case. If the class is superstable, one of the consequences of not having the
appropriate order property is that (see 4.15) for a strictly indiscernible sequence
a
t
: t I) over A each a
t
of length at most and

b, singleton for simplicity,
for all except few of the t, a
t
b realizes the same type (= convergence, existence

of average). Of course, we can get better theorems generalizing the ones for rst
order theories: we can use /
(C) and/or demand that after adding to A, c and

few of the a
t
s the rest is strictly indiscernible over the new A, but this is not used
in 5.5. Now if C is (, 1)-superstable the number of exceptions is nite, however,
the inverse is not true: for some non (, 1)-superstable class C still the number of
exceptions in such situations is nite. In the proof of 5.5(1) this property is used
as a dividing line.
Proof of 5.5. 1)
Case I There are M, N, p, a
i
: i < i
) as in 4.13()
and c, (in fact g( c) = 1) such

that c realizes h
i
(p) for innitely many is and fails to realize h
i
(p) for innitely
many is.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

Let I be a ( +
(2
)
+)
+
-strongly saturated dense linear order (see Denition
4.18) such that even if we omit
(2
)
+ members, it remains so. By the strict
indiscernibility we can nd a
t
: t I), c as above.
So there is u I, [u[ <
(2
)
+ such that q = tp( a
t
c, , C) is the same for all
t Iu; without loss of generality q = tp( a
t
c, , C) t Iu, so u is innite.
Hence we can nd i
n
i
u such that i
n
< i
n+1
. Let I
= I(ui
n
: n < ),
so that I
is still
+
-strongly saturated. Hence for every J I
of order type for

some c
J
( C) we have
t I
J tp( a
t
c
J
, , C) = q
t J tp( a
t
c
J
, , C) ,= q.
This clearly suces.
Case II Not Case I.
As in [Sh 3] (the nitely many nite exceptions do not matter) or see part (2).
2) If < 2
the conclusion follows from 3.3(2). Possibly decreasing (allowable as

(K) rather than
(K) is assumed) we can nd a tree T

, so closed
under initial segments such that [T
>
[ but [T
[ > . (The assumption
> LS(K) is needed just for this). Let M

i
: i +1), c M
+1
exemplify

(K) and let T
= T 0) :

Ord and i < i T . Now we
can by induction on i + 1 choose h
: T
i
), such that:
(a) h
is a
K
-embedding from M
g()
into C
(b) j < g() h
j
h
(c) if i = j+1, T
j
, then h
(M
i
) : Suc
T
()) is strictly indiscernible,
and can be extended to a sequence of length such that h
(p M
i
) :
Suc
I
()) is contradictory (i.e. as in Denition 4.9(1)).
There is no problem to do this. Let M
K
C be of cardinality and include
(M
i
) : i < and T
i
hence it includes also h
(M
) if T
as
M
=
_
i<
M
i
.
For T

let c
= h
<0>
(c) and M
= h
(M
i
) when T
i
Ord and
i + 1, so by 4.15 clearly (by clause (C))
() if i < , T
i
, and
1
T
, then
Suc
T
() : for some
1
,
1
T
and
c
1
realizes tp(c
1
, h
1
(i+1)
(m
i+1
))
has cardinality <
(2
+LS(K)
)
+.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

30 SAHARON SHELAH
Next dene an equivalence relation e on T
1
e
2
i tp(c
1
, M) = tp(c
2
, M).
or just
1
e
2
i ()[ T tp(c
1
, M
) = tp(c
2
, M
)].
Now if for some T
, [/e[ >
(2
+LS(K)
)
+ then for some
T
>
, we
have
(g(
+ 1)) : /e has cardinality >

(2
+LS(K)
)
+
which contradicts (); so if
(2
+LS(K)
)
+, we are done.
But if for some T
>
the set in () has cardinality , then we can
continue as in case I of the proof of part (1) replacing innite by of cardinality
, so assume this never happens. So above if [/e[ > 2
, we get again a
contradiction. So if [T
[ > 2
, we conclude [T
/e[ = [T
[, so we are
done. We are left with the case < 2
, covered in the beginning (note that for

< 2
the interesting notion is splitting).

5.5
5.7 Claim. If > , LS(K, K) and K is categorical in and ,=
+
, then
1) K is (, 1)-superstable.
2)
(K) is empty.
Proof. 1) Assume the conclusion fails. If >
+
, we can use 5.5 + 1.7 so
without loss of generality < <
+
hence cf() > LS(K).
By clause (b) of 1.7 if M K
then M is cf()-saturated. On the other hand

from the Denition of (, 1)-superstable we shall get below a non-
+
-saturated
model.
Let =
(2
)
+. Assume K is not (, 1)- superstable so we can nd in K
an increasing continuous sequence M

i
: i + 1) and c M
+1
such that
p
n+1
= tp(c, M
n+1
, M
+1
) -strongly splits over M
n
for n < . For each n <
let a
n
i
: i < ) be a strictly indiscernible sequence over M
n
exemplifying p
n+1
does -strongly split over M
n
(see Denition 4.12). So we can dene a
n
i
C for
i [, ) such that a
n
i
: i < ) is strictly indiscernible over M
n
. Let T
n
=
2n
: (2m) < (2m+ 1) for m < n. For n < , i < j < let h
n
i,j
AUT(C) be
such that h
n
i,j
M
n
= id
M
n
, h
n
i,j
( a
n
0
a
n
1
) = a
n
i
a
n
j
. Now we choose by induction on
n < , f
: T
n
), g
: T
n
), a
i
: i < , T
n
) such that:
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

(a) f
are restrictions of automorphisms of C

(b) Dom(f
) = M
n
(c) g
AUT(C)
(d) a
i
= g
( a
n
i
) if T
n
(e) f
<>
= id
M
0
,
(f) f
(g) if
2n
, m < n then f
(2m)
f
(h) if
2n
and i < j < then f
<i,j>
(g
h
n
i,j
) M
n+1
.
There is no problem to carry the induction. Now choose by induction on n, M
n
,
n
, i
n
, j
n
such that
() i
n
< j
n
< and
n
= i
0
, j
0
, . . . , i
n1
, j
n1
) so
n
T
n
() M
n
K
, M
n

K
M
n+1
() Rang(f
n
) M
n
() a
n
i
n
, a
n
j
n
realizes the same type over M
n
() a
n
i
n
, a
n
j
n
M
n+1
.
There is no problem to carry the induction (using the theorem on existence of
strictly indiscernibles to choose i
n
< j
n
).
So
_
n<
f
n
can be extended to f AUT(C). Let c
= f(c), M
=
_
n
M
n
, M
+1
<
K
C includes M
f(M
+1
). Clearly tp(c, M
n+1
, M
+1
) does -split over M
n
hence
M
is not
+
-saturated (as cf() > ) (see 5.8 below); contradiction.
2) Similar proof.
5.7
5.8 Claim. If LS(K), M
i
: i ) is
K
-increasing continuous,
p S
(M
), p -strongly splits over M

i
for all i (or just -splits over M
i
) and
<
+
then M
is not
+
-saturated.
Proof. Straight.
5.9 Claim. Assume there is a Ramsey cardinal > + LS(K). If K is not (, 1)-
superstable, then for every > + LS(K) there are 2
pairwise non-isomorphic
models in K
.
Proof. By [GrSh 238] for regular; together with [Sh:e] for all .
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

32 SAHARON SHELAH
5.10 Lemma. 1) If for some M, [S(M)/E
[ > |M|+
(2
)
+ and LS(K)
then K is not (, 1)-superstable.
2) If
[S(M)/E
[ >
<
|M|+
(2
)
+, LS(K) + then
(K).
Proof. No new point when you remember the denition of E
(see 1.8).
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

6 Existence of Many Non-Splitting
Below alternatively we can start with 6.8.
6.1 Question. Suppose + LS(K) < and

N = N
i
: i ) is <
1
,
-
increasing continuous (we mean for i < j, j non-limit N
i
<
1
,
N
j
), <
+
(a limit
ordinal) and p S
m
(N
). Is there < such that for every M K
, N

K
M, p
has an extension q S
m
(M) which does not -split over N
(and so in particular
p does not -split over N
)?
6.2 Observation. Let , , ,

N and M be as in 6.1.
1) If p N
+1
does not -split over N
, then p N
+1
has at most one extension
in S(M) mod E
which does not -split over N
because N
+1
K
is universal
over N
, N
+1

K
M K
. So in 6.1 if p does not -split over N
, then there is
at most one q/E
for q as there.
2) If the asnwer is yes and p, q S(N
) and i < p N
i
= q N
i
then p = q.
3) If the answer is yes and p does not split over N
and N

K
M K
then
(i) there is q S
m
(M) which does not -split over N
and q N
+1
= p
N
+1
(ii) this q is unique and satises p = q N
.
Proof. E.g.,
2) For some i
1
< , p does not -split over N
i
1
and there is i
2
< , q does not -split
over N
i
2
. By monotonicity of non--splitting, without loss of generality i
1
= i = i
2
.
Let a be a sequence of length listing N
, and let f Aut(C) extends id

N
i
and
maps N
into N
i+1
and let a
= f( a).
Now if c
1
, c
2
C realizes p, q respectively then tp(c
) a, N
i
, C) = tp(c
)( a
, N
i
, C)
for = 1, 2 as p, q does not -split over N
i
, tp(c
1
) a
, N
i
, C) = tp(c
2
) a
, N
i
, C) as
p N
i+1
= q N
i+1
. Together tp(c
1
) a, N
i
, C) = tp(c
2
) a, N
i
, C) which means
tp(c, N
, C) = tp(c
2
, N
, C).
6.2
6.3 Lemma. Suppose K is categorical in , cf() > LS(K). Then the answer
to question 6.1 is yes.
6.4 Remark. We intend later to deal with the case > cf() + LS(K) as in
[KlSh 362].
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

34 SAHARON SHELAH
Notation. I is I +I +. . . ( times) (with the obvious meaning).
Proof. Let be proper for linear order such that [()[ LS(K), EM
(K)
(I, )
K (of cardinality [I[ + (K)) where I is a linear order, of course and I J
EM
(K)
(I, )
K
EM
(K)
(J, ). Let I
be a linear order of cardinality such

that I
(+1)

= I
for <
+
and I

= I
and I
[= a < b implies that I
is
isomorphic to I
(a, b), see [Sh:e, AP,2]. By 1.7 we know that EM

(K)
(I
, )
is
+
-saturated.
First assume only N
i
<
0
,
N
i+1
for i < ; (or just N
i+1
is universal over N
i
).
Now we choose by induction on i a triple (
i
, N
i
, h
i
) for i
(a)
i
is an ordinal <
+
, increasing continuous with i
(b) N
i
K
is
K
-increasing continuous with i
(c) h
i
is an isomorphism from N
i
onto N
i
, increasing continuous with i
such that
(d) N
0

K
EM
(K)
(I
i
, )
(e) EM
(K)
(I
i
, )
K
N
i
EM
(K)
(I
i+1
, )
(f) if i is a limit ordinal then N
i
= EM
(K)
(I
i
, ).
For i = 0, as EM(I
, ) is
+
-saturated there is a K-embedding h
0
of N
0
into
EM
(K)
(I
, ). As Rang(h) has cardinality , there is u

0
of cardinality
such that Rang(h
0
) EM
(K)
(I
u
0
, ). So
0
=: otp(u
0
) is an ordinal
[
i
,
i+1
) hence EM
(K)
(I

0
, )

= EM
(K)
(I

i
, ) so without loss of
generality u
0
=
i
.
For i limit take union. The case i = j +1 is similar to i = 0 using amalgamation.
As we have used only N
i+1
universal over N
i
by replacing N
i
: i ) by a
longer sequence and renaming without loss of generality N
i
= EM
(K)
(I
i
, ).
Alternatively,
1
if <
+
then for some (,
+
) the model EM
(K)
(I
, ) is
K
-
universal over EM
(K)
(I
, ).
[Why? We know that there is N K
M
universal over EM
(K)
(I
, ). As
EM
(K)
(I
, ) is
+
-saturated there is a
K
-embedding g of N over EM
(K)
(I
, ) into EM
(K)
(I
, ). As [Rang(g)[ there is a set u of cardinality

which includes and Rang(N) EM
(K)
(I
u, ).
So otp(u, <) is an ordinal of cardinality call it and let h : u be an
isomorphism, so h = id
and let

h be the isomorphism from EM(I
u, )
onto EM(I
, ) which h induces. Clearly it is the identity on EM(I , ).

Now ,
h g are as required.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

2
if <
+
and then for some (,
+
), EM
(K)
(I
, ) <
1
,
EM
(K)
(I
, ) is ??
[Why? Iterate
1
]
3
there are
i
: i ), an increasing continuous sequence of ordinals <
+
and h
i
: i ) such that h
1+i
is an isomorphism from N
i
onto EM
(K)
(I
i
, ), h
i
increases with i.
[Why? By
2
and the uniqueness for <
1
,
.]
Now h
is dened h
: N
onto
EM
(K)
(I
, ), so as EM
(K)
(I
, )
is
+
-saturated, h
(p) is realized in EM
(K)
(I
, ) say by a, so let a =
(x
(t
1
,
1
)
, . . . , x
(t
n
,
n
)
) where is a sequence of terms in () and (t
) is in-
creasing with (in I
). Let < be such that:

_
1
, . . . ,
n
_

.
Let
=
_

if
<
if
Then in the model N = EM

(K)
(I
+, ), we shall show that the nite sequence

a
=
_
x
(t
1
,
1
)
, . . . , x
(t
n
n
)
_
realizes a type as required over M = EM
(K)
(I
, ).
Why? Let M
= EM
(K)
(I
, ) for < . Assume toward contradiction that

() tp( a
, M, N) does -split over M

+1
.
Let c,
b

M realize the same type over M
+1
but witness splitting.
We can nd w , [w[ such that c,
b EM(I
w, ). Choose such
that
sup(w) < < .
Let M
= EM
(K)
(I
w [, )), ) <
K
M.
Let N
= EM
(K)
(I
w [, ) [, +)), ) <
K
N.
So still c,
b witness that tp( a
, M
, N

+1
.
There is an automorphism f of the linear order I
w [, )) [, +))
such that
f (I
+1
) = the identity
f (I
[ + 1, +)) = the identity

(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

36 SAHARON SHELAH
Rang
_
f (I
w)
_
I
[
+1
,
+2
).
Now f induces an automorphism of N
naturally called

f.
So
f M
= identity
f( a
) = a
f(M
) = M
As

f is an automorphism,

f(c),

f(
b) witness that tp(

f( a
),

f(M
),

f(N
)) does -
split over

f(M
+1
); i.e. tp( a
, M
, N
+1
. So tp( a
, M
+2
, N)
does -split over M
+1
.
Now choose
<
+
for (,
+
], increasing continuous by
+i
=
+i
M
= EM
(K)
(I
, ).
So M
: ) is increasing continuous. So for

1
[,
+
) there is f AUT(I
( +)) such that

f I
= identity
f takes I
,
+1
) onto I
1
+1
)
f takes I
[
+1
,
+2
) onto I
1
+1
f takes I
[
+2
,
1
+2
) onto I
1
+2
f I
1
+2
, +) = identity.
As before this shows (using obvious monotonicity of -splitting)
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

tp( a
, M
1
+2
N) -splits over M
1
+1
.
So < : tp( a
, M
+1
, N) does -split over M
has order type , so without

loss of generality is . By 3.3(2) we get a contradiction.
6.3
6.5 Theorem. Suppose K categorical in and the model in K
is
+
-saturated
(e.g. cf() > ) and LS(K) < < .
1) M <
1
,
N N is saturated if LS(K) < .
2) If
1
,
2
are regular cardinals (or just limit ordinals) and for = 1, 2 we
have M
<
1
,
, then N
1

= N
2
.
3) There is M K
which is saturated.
4) If
1
,
2
are as above M
1
,
N
1

=
M
N
2
(in fact this holds for =
LS(K), too).
6.6 Remark. 1) The model we get by (2) we call the saturated model of K
in .
2) Formally - we do not use 6.3.
3) If M <
1
,
N, we call N brimmed over M.
Proof. 1) By the uniqueness proofs 2.2 as M <
1
,
N there is an <
K
-increasing
continuous sequence M
i
: i ) satisfying M
i
<
1
,
M
i+1
, M
0
= M, M
= N
and as in the proof of 6.3 without loss of generality M
i
= EM
(K)
(
i
, ) where
i
<
+
.
To prove N = N
is -saturated suppose p S
1
(M
), M
K
N, |M
| < ;
as we can extend M
(as long as its power is < and it is <

K
N), without loss of
generality M
= EM
(K)
(J, ), J
, [J[ < .
So for some we have [, + ) J = and +
. We can replace
[, +) by a copy of ; this will make the model -saturated. That is we can nd
a linear order I
such that (
, <) I
and t I
[= < t < ( + 1)
and = otp(I
+
t : t I
) so EM
(K)
(I
, ) is a
K
-extension of N = N
and belongs to K
hence is
+
-saturated [alternatively, use I
ordinal as in a
proof of 6.3].
But easily this introduces no new types realized over M
. So p is realized.
2) In detail assume M <
,
for = 1, 2. So we can nd M
,i
: i ) is
K
-increasing continuous such that M
K
M
,0
, M
,i
<
1
,
0
M
,i+1
and M
,
= N
.
So let
,i
< be increasing continuous for i
, divisible by and an isomor-

phism h
from N
onto EM
(K)
(
,
, ) such that h
1
M = h
2
M, h
(N
,i+1
) =
EM
(K)
(
,1+i
, ). Let
1,
= I
j
: j <
2
, I
j
increasing continuous with
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

38 SAHARON SHELAH
j, [I
j+1
I
j
[ = . Easily EM(I
j
, ) : j ) exemplify that h
1
(N
1
) is (,
2
)-
brimmed over h(M). So M <
1
,
1
N M <
1
,
1
N and so by 2.2(5) we are done.
3) Follows from the proof of part 1) and 2.2(1) + 1.7.
4) Similarly.
6.5
Remark. In part (1) we have used just cf() > > LS(K).
6.7 Claim. 1) Assume K categorical in , cf() > > LS(K). If N
i
K
is saturated, increasing with i for i < and <

+
then N =
_
i<
N
i
K
is
saturated.
2) [K categorical in , cf() > > LS(K)]. Possibly changing (actually as in
Denition 8.4(2). If I is a linear order of cardinality then EM
(K)
(I, ) is -
saturated. Moreover, if I J are linear order of cardinality then EM
(K)
(J, )
is a universal
K
-extension of EM
(K)
(I, ) and even brimmed over it.
Proof. 1) We prove this by induction on , so by the induction hypothesis without
loss of generality N
i
: i < ) is not just
K
-increasing and contradicts the con-
clusion but also is increasing continuous and each N
i
saturated. Without loss
of generality = cf(). If cf() = the conclusion clearly holds so assume
cf() < . Let M
K
N, |M| < and p S(M) be omitted in N and let
= + |M| + LS(K) < , and let p q S(N). Now we can choose by in-
duction on i , M
i

K
N
i
and M
+
i

K
N such that M
i
K
, M
+
i
K
, M
i
is
K
-increasing continuous and M N
i
M
i
, j < i M
+
j
N
i
M
i+1
and
M
i
<
1
,
M
i+1
and if q does -split over M
i
then q M
+
i
does -split over M
i
.
So by 6.3, 6.5 we know that M
is saturated, and for some i() < we have:

q M
does not -split over M

i()
. But M
+
i()
N =
_
i<
N
i
, M
+
i()
N
j
M
j+1
so M
+
i()
M
. So necessarily q S(N) satises i() i < implies that q N

i
i()
.
Now we choose by induction on <
+
, M
i(),
, b
, f
such that:
M
i(),
K
, M
i()

K
M
i(),

K
N
i()
, M
i(),
is
K
-increasing continuous in
, b
N
i()
realizes q M
i(),
, f
is a function with domain M
and range
N
i()
such that the sequences c = c : c M
) and c
=: f
(c) : c M
) realize
the same type over M
i(),
and b
Rang(f
) M
i(),+1
. As N
i()
K
is
saturated and LS(K) < we can carry the construction; if some b
realizes
q M
we are done (as p = q M, M

K
M
and b
N realizes p). Let d C

realize q so
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

()
1
< <
+
c
) does not realize tp( cd), M

i()
, C).
[Why? As cb
) does not realize tp( cd), M

i()
, C) because d realizes p =
q c whereas b
does not realize p = q c.]

On the other hand as q does not -split over M
i()
we have
tp( cd), M
i()
, C) = tp( c
d), M
i()
, C) so by the choice of b
:
()
2
if < <
+
then c
) realizes tp( cd), M

i()
, C).
We are almost done by 4.15.
[Why only almost? We would like to use the -order property fail, now if we could
dene c
) : for < (2
)
+
) ne, but we have only <
+
, this is too short.]
Now we will rene the construction to make c
) : <
+
) strictly indiscernible
which will be enough. As N
i()
is saturated without loss of generality N
i()
=
EM
(K)
(, ) and M
i()
= EM
(K)
(, ) (using 6.8 below). As before for some
<
+
there are sequences c
in EM
(K)
( + , ) realizing tp( c, N
i()
, C), q
N
i()
respectively, here we use cf() > rather than just cf() . For each
<
+
there is a canonical isomorphism g
from EM
()
( [, + ), ) onto
EM
()
( +, ). So without loss of generality M
i(),
= EM
(K)
( +
, ), c
=
g
+
( c
), b
= g
+
(b
). So ()
1
+ ()
2
gives the (, 1, )-order property contra-
dicting categoricity by 4.8(1) as
+
< .
2) By (1) and 6.8 below.
6.7
We really proved, in 6.5 (from categoricity):
6.8 Subfact. Assume K is categorical in .
1) If I J are linear order, of power < cf();
() t JI
_
0
s J
_
[s
I
t] where s
I
t means s, t realize the same
Dedekind cut,
then every type over EM
(K)
(I, ) is realized in EM
(K)
(J, ).
2) If I J are linear orders of cardinality < cf(), , [JI[ = [J[ and () above
then EM
(K)
(I, ) <
1
,
EM
(K)
(J, ) where.
3) We can nd
(in fact

or
LS(K)
in the notation of 8.3(3)) such that
(a)
is as in 0.6
(b) if I J and LS(K) [I[ [J[ then EM
(K)
(J,
) is a brimmed over
EM
(K)
(I,
).
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

40 SAHARON SHELAH
Proof. 1) Why? Use the proof of 6.5(1).
Replace the cut of t in I by : we get cf()-saturated model.
2) As in the proof of 6.5(2).
3) For a linear order I we can dene I
= : is a nite sequence, and for

i < g() we have i even (i) I and i odd (i) 1, 1.
Ordered by <
I
i
( even i)( i = i (i) <
I
(i)) or
( odd i)( i = i (i) = 1 (i) = 1) or
g() odd g() < g() (g()) = 1 or
g() even g() < g().
We can choose
, [(
)[ = LS(K) such that for every linear oder I, EM(I,
) =
EM(I
, ). Now if I J, [LS(K) [J[ then [J
[ = J, I
, moreover for every

t J
the set s J
: s realizes the same cut of I
as t has cardinality [J[;

moreover we can nd [J[ pairwise disjoint intervals of J
, disjoint to I
so list them
as (a
, b
) : < [J[ [J[) for i < [J[ [J[. Let I

0
= I, I
1+i
= J
[a
, b
[ : i
< [J[ [J[ for i < [J[ [J[. So I
: [J[ [J[) is an increasing continuous

sequence of suborder of J
, with I
0
= I
, I
|J||J|
= J
. Let M
= EM(I
, ).
So M
: [J[ [J[) is increasing continuous, M

0
= EM(I,
), M
|J||J|
=
EM(J,
), M
= M
(K) : [J[ [J[) is

K
-increasing continuous. By the
previous parts (and the choice of the I
s), every p S(M
) is realized in M
+1
,
hence M
|J||J|
= EM
(K)
(J,
) is brimmed over M
0
= EM
(K)
(I, ) as required.
6.8
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

7 More on Splitting
7.1 Hypothesis. As before + conclusions of 6 for [LS(K), cf()).
So
()(a) K has a saturated model in .
(b) union of increasing chain of saturated models in K
of length
is saturated.
(c) if M
i
: i ) increasing continuous in K
, each M
i+1
saturated over M
i
(the previous one), p S(M
) then for some i < , p does not -split over

M
i
.
7.2 Conclusion. If p S
m
(M) and M K
is saturated, then for some

M
<
1
,
M, M
is saturated and p does not -split over M
.
Proof. We can nd M
n
: n ) in K
, each M
n
saturated M
n

1
,
M
n+1
and
M
=
_
n<
M
n
so as M
is saturated, without loss of generality M
= M. Now
using ()(c) of 7.1 some M
n
is O.K. as M
.
7.2
7.3 Fact. If M
0

1
,
M
2

1
,
M
3
, p S
m
(M
3
) and p does not -split over M
0
,
then R(p) = R(p M
2
), see Denition 5.1.
Proof. We can nd (by uniqueness) M
1
K
such that M
0

1
,
M
1

1
,
M
2
and
we can nd M
4
K
such that M
3

1
,
M
4
.
We can nd an isomorphism h
1
from M
3
onto M
2
over M
1
(by the uniqueness
properties <
1
,
). By uniqueness 2.2(1) there is an automorphism h of M
4
extending
h
1
. Also by uniqueness there is q S(M
4
) which does not -split over M
0
and
extend p M
1
(e.g., there is an isomorphism g from M
3
onto M
4
over M
2
and let
q = g(p)). As p, q M
3
do not -split over M
0
and have the same restriction to
M
1
and M
0

1
,
M
1
clearly p = q M
3
. Consider q and h(q) both from S(M
4
),
both do not -split over M
0
and have the same restriction to M
1
and id
M
1
h; as
M
0
<
1
,
M
1
it follows that q = h(q).
So R(p M
2
) = R(q M
2
) = R(h(q M
3
)) = R(q M
3
) = R(p) as required.
7.3
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

42 SAHARON SHELAH
7.4 Claim. [Here?] [K categorical in , cf() > > LS(K)].
Suppose m < , M K
is saturated, p S
m
(M), M
K
N K
, p q
S
m
(N), N brimmed over M, q not a stationarization of p (i.e. for no M
<
,
M,
q does not -split over M
). Then q does -divide over M.

Proof. By 7.5 below and 6.3 (just p does not -split over some N
m
where
N
: ) witness N
0
<
1
,
M).
7.5 Claim. [K categorical in and cf() > > LS(K)]
Assume M
0
<
1
,
M
1
<
1
,
M
2
all saturated. If q S(M
2
) does not -split over
M
1
and q M
1
0
, then q does not -split over M
0
.
Proof. Let M
3
K
be such that M
2
<
1
,
M
3
and c M
3
realizes q. Choose a
linear order I
of cardinality such that I
( +
)

= I

= I
, remember
that on the product we do not use lexicographic order. I
has no rst nor last

element (see [Sh 220, AP]).
Let I
0
= I
, I
1
= I
0
+ I
Z, I
2
= I
1
+ I
Z, I
3
= I
2
+ I
.
Clearly without loss of generality M
= EM
(K)
(, I
), let c = ( a
t
0
, . . . , a
t
k
) so
t
0
, . . . , t
k
I
3
; let I
1,n
= I
0
+ I
m : Z [= m < n and I
2,n
= I
1
+ I
m :
Z [= m < n and I
0,
= I
. So we can nd a (negative) integer n() small

enough and m() Z large enough such that t
0
, . . . , t
n
I
2,n()+1
I
1,m()1
. Let
M
1,n
= EM
(K)
(I
1,n
, ) and M
2,n
= EM
(K)
(I
2,n
, ). Clearly M
0
<
1
,
M
1,n
<
1
,
M
1
<
1
,
M
2,n
<
1
,
M
2
. Clearly (use automorphism of I
3
)
()
0
q M
2,n
1,m
if Z [= n < n(), m() m Z.
As q does not -split over M
1
and M
2,n+1
is brimmed over M
2,n
for n Z, etc.,
by 7.3 with q, M
1
, M
2,n
, M
2
, q here standing for M
0
, M
2
, M
3
, p there we get
()
1
R(q) = R(q M
2,n
) if n Z.
Similarly
()
2
R(q M
1
) = R(q M
1,m
) if m Z.
By ()
0
and 7.3 we have
()
3
R(q M
2,n()
) = R(q M
1,m()
).
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

Similarly we can nd a successor ordinal () < and k() Z such that
t
0
, . . . , t
k
I
1,k()+1
I
0,()1
and then prove
()
4
R(q M
0
) = R(q M
0,
) if () <
()
5
R(q M
1,()
) = R(q M
0,
) if () < .
Together R(q) = R(q M
0
), hence q does not -split over M
0
as required.
7.4
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

44 SAHARON SHELAH
PART II
8 Existence of nice
We build EM models, where equality of types over A in the sense of the existence
of automorphisms over A behaves nicely.
8.1 Context.
(a) K is an abstract elementary class with models of cardinality
(2
LS(K)
)
+.
8.2 Remark. Mostly it suces to assume ((), () for 8.6, 8.7 omitting the second
clause in 8.6(b), 8.7(3); () () for
(a)
() K is a class of (K)-models, which is PC

2
,
, and
() we interpret LS(K) as such that
() K has a model of cardinality
(2
LS(K)
)
+
()
K
is a PC
2
,
partial order
K
on K
()
K
is closed under increasing continuous chains (in 8.5(3) hence
or
,= (see below) for

LS(K) is not empty)
() preserve indiscernible isomorphism.
8.3 Denition. 1) Let LS(K), now
or
=
or
[K] is the family of proper

for linear orders (see [Sh:c, Ch.VII]) such that:
(a) [()[ and
K
()
(b) EM
(K)
(I, ) = EM(I, ) (K) K
(c) I J EM
(K)
(I, )
K
EM
(K)
(J, )
(d) is as in 6.8(3), (needed only in 9).
2)
or
=
or
[K]
is
or
LS(K)
.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

8.4 Denition. We dene partial orders
and
on
or
(for LS(K)):
1)
1

2
if (
1
) (
2
) and EM
(K)
(I,
1
)
K
EM
(K)
(I,
2
) and EM(I,
1
) =
EM
(
1
)
(I,
1
) EM
(
1
)
(I,
2
) for any linear order I.
Again for = LS(K) we may drop the .
2) For
1
,
2

or
, we say
2
is an inessential extension of
1
and write
1

ie

2
if
1

2
and for every linear order I, we have
EM
(K)
(I,
1
) = EM
(K)
(I,
2
).
(note: there may be more functions in (
2
)!)
3) Let
lin
be the class of proper for linear order and producing linear orders
such that:
(a) () has cardinality ,
(b) EM(I, ) is a linear order which is an extension of I: in fact
[t I x
t
= t].
4)
1

2
i there is such that
(a)
lin
(b)

or
for = 1, 2
(c)
2

ie

2
where
2
=
1
, i.e.
EM(I,
2
) = EM(EM(I, ),
1
).
(So we allow further expansion by functions denable from earlier ones (composition
or even denition by cases), as long as the number is ).
8.5 Claim. 1) (
or
) and (
or
) are partial orders (and
).
2) If
i
: i < ) is a
-increasing sequence, <

+
, then it has a <
-l.u.b. ;
EM(I, ) =
_
i<
EM(I,
i
).
3) Similarly for <
.
Proof. Easy.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

46 SAHARON SHELAH
8.6 Lemma. [The a.e.c. omitting type theorem]
If N
K
M, |M|
(2
)
+, |N| +LS(K), then there is
or
so is proper
for linear order such that:
(a) EM
(K)
(, ) = N
(b) N
K
EM
(K)
(I, ), and recall
I J EM
(K)
(I, )
K
EM
(K)
(J, )
(c) EM
(K)
(I, ) omits every type p S(N) which M omits, moreover if I is
nite then EM
(K)
(I, ) can be
K
-embedded into M
(d) [
[ |N| + LS(K) and non-trivial, hence, [EM(I, )[ = [I[ + [
[ for
every linear order I.
Proof. This is a particular case of 8.7 below when N
1
= N
0
; which is proved in
details (or see straight by [Sh 88, 1.7] or deduce by 4.6).
8.6
8.7 Lemma. Assume
(a) LS(K)
(b) N
0

K
N
1

K
M
(c) |N
0
| , |N
1
| = and |M|
(2
)
+()
(d)
0
= p
0
i
: i < i
0
S(N
0
) each p
0
i
omitted by M
(e)
1
= p
1
i
: i < i
1
S(N
1
) such that for no i < i
1
and c M
does c realize p
1
i
/E
[see Denition 1.8; where c realizes p

1
i
/E
means that
c realizes every restriction p
1
i
M, M
K
N
1
, M K
].
Then we can nd N
: ), and q
1
i
: i < i
1
) such that
() proper for linear orders
() N
is
K
- increasing continuous (for )
() N
0
= N
0
and N

K
N
1
() q
1
i
S(N
) and q
1
i
= p
1
i
N
() EM
(K)
(, ) is N
0
() for linear order I J we have
EM
(K)
(I, )
K
EM
(K)
(J, )
() for each
6
nite linear order I, there is a
K
-embedding h
I
of EM
(K)
(I, )
into M which extends id
N
|J|
6
the price for this nice formulation is that it may fail to satisfy EM(I
1
, ) EM(I
2
, ) =
EM(I
1
I
2
, ) for I
1
I
2
I
1
, i.e., for some n-place function f(x
t
1
, . . . , x
t
n
) may be even
constant.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

() [main clause] for any linear order I, EM
(K)
(I, ) omits every p
0
i
for i < i
0
and omits every q
1
i
in a strong sense: for every a EM
(K)
(I, ) and nite
J I such that a EM
(K)
(J, ) we have q
1
i
N
|J|
,= tp(a, N
|J|
, h
J
(EM
(K)
(I, ))).
8.8 Remark. 1) So we really can replace q
1
i
by q
1
i
N
n
: n < ), but for -chains
by chasing arrows such limit (q
1
i
) exists, see 1.12.
2) If a is a sequence in a model M, c( a, M) is the closure of Rang( a) under the
functions of M.
Proof. By [Sh 88, 1.7] (and see 0.5) we can nd
1
, (K)
1
, [
1
[ (here we can
have [
1
[ LS(K) ) and an expansion M
+
of M to a
1
-model and a set of
quantier free types (so [[ 2
0
+|LS(K)|
) such that:
(A) () M
+
omits every p
() if M
is a
1
-model omitting every p then M
(K) K
and N
(K)
K
M
(K).
So
(B) for a
>
M we let M
+
a
= M
+
c( a, M
+
) then M
+
a
(K)
K
M
+
(K),
Rang( a) Rang(
b) M
+
a
(K)
K
M
+
b
(K) where
a
>
(N
) [M
+
a
[ N
.
Note that M
+
a
has always cardinality . Note that further expansion of M
+
to
M
, as long as [(M
)[ preserves (A) + (B); so we can add (for clause (E) we

use the assumption (e), i.e., M omits p
1
i
/E
, not just p
1
i
)
(C) N
0
, M
+
have the same universe

and let M
+
a,1
= M
+
a
([N
1
[ [M
+
a
[)
(D) N
0

K
M
+
a,1
(K)
K
M
+
a
(K)
(E) for i < i
1
, the type p
1
i
(M
+
a,1
(K)) is not realized in M
+
a
(K);
(F) N
0
, N
1
are closed under the functions of M
+
, so N
+
0
= M
+
[N
0
[, N
+
1
=
M
+
(N
1
) are well dened
1
-models omitting every p .
Now we choose by induction on n, sequence f
n
: < (2
)
+
) and N
n
such that:
(i) f
n
is a one-to-one function from
() into M
(ii) f
n
() : <
()) is n-indiscernible in M
+
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

48 SAHARON SHELAH
(iii) moreover, if , < (2
)
+
, and m n and
1
< . . . <
m
<
() and
1
< . . . <
m
<
() then: the sequences a = f

n
(
1
), . . . , f
n
(
m
)),
b = f
m
(
1
), . . . , f
m
(
m
)) realize the same quantier free type in M
+
over
N
+
1
, so there is a natural isomorphism g
b, a
from M
+
a
onto M
+
b
(mapping
f
) to f
)), moreover
(iv) i < i
1
g
b, a
(p
1
i
(M
+
a,1
(K))) = p
1
i
(M
+
b,1
(K))
(v) if m < n then for every < (2
)
+
there are
satisfying <
< (2
)
+
and h
an increasing function from
() to
() such that <
()
f
n
() = f
m
(h
())
(vi) N
n

K
M
1
and for every m n < , < (2
)
+
and
0
< . . . <
m
1
<
(), N
n
is (M
+
1
M
+
f
n
):<m
) (K) = M
f
n
(
n
):<m,1
.
As the indiscernibles in clause (iii) are over N
+
1
we can dene, for n 1, N
g( a)
=
(N
+
1
M
+
a,1
) (K) for any a as in (iii), i.e., this restriction does not depend on n.
For n = 0 this is trivial. The induction step n + 1 rst for each < (2
)
+
we
apply Erdos Rado theorem for f
n
+
getting Y
n

+
() of cardinality
()
as in (iii) (also for (iv)) for = . Then by the pigeon hole principle for some
X
n
(2
)
+
of cardinality (2
)
+
we have: if
1
,
2
X
n
,
1
<
n
belongs to Y
for = 1, 2 then f
n
1
+
(
1
1
), . . . , f
n
1
+
(
1
n
)), f
n
2
+
(
2
1
), . . . , f
n
2
+
(
2
n
)) realize the
same quantier free type in M
+
over N
+
1
.
Now we choose f
n+1
() as f
n
+
(
) where
X, otp(
X) = ,
,
otp(
) = . Having carried the induction we choose such that for every n <
,EM(n, ) is isomorphic to M
+
a
whenever a = f
n
) : < n, by an isomorphic
mapping a
to f
n
) for < n wherever < (2
)
+
,
0
< . . . <
n1
<
().
8.7
8.9 Denition. 1) Let K
or(+)
be the class of I a linear order expanded by the
unary relations P
I
1
, P
I
2
such that P
I
1
is an initial segment of I and P
I
2
= IP
I
1
. Let
() be the vocabulary <, P
1
, P
2
.
2) For LS(K) let
or(+)
=
or(+)
[K] be the family of proper for K

or(+)
(see
[Sh:c, Ch.VII]), such that
(a) () extends
K
and has cardinality
(b) for every I K
or(+)
, EM(I, ) is a ()-model which is the closure (by the
functions F
M
, F () a function symbol) of the skeleton a
t
: t I); for
simplicity a
t
= a
t
) and, of course, s ,= t a
s
,= a
t
; and let EM
(I, ) be
the -reduct of EM(I, ) for ()
(c) a
t
: t I) is qf-indiscernible in EM(I, ) which means that: if t
0
<
I
. . . <
I
t
n+1
, s
0
<
I
. . . <
I
s
n1
and s
P
I
1
t
P
I
1
for < n then
a
t
: < n), a
s
: < n) realizes the same quantier free type in EM(I, )

(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

(d) if I J K
or(+)
then EM(I, ) EM(J, ) and EM
(K)
(I, )
K
EM
(K)
(J, ), so both are in K.
3) If = LS(K) we may omit it.
4) If I K
or(+)
,
or(+)
and h is a partial automorphism of I, then

h is
the following function; letting I
0
= Dom(h) we have: if n < , t
0
<
I
. . . <
I
t
n1
are from I
0
and (x
0
, . . . , x
n1
) is a ()-term then

h((a
t
0
, . . . , a
t
n1
) =
(a
h(t
0
)
, . . . , a
h(t
n1
)
).
8.10 Denition. 1) Let
lin(+)
be the class of proper for K

or(+)
such that
has cardinality , the two-place relation < and unary predicates P

1
, P
2
belong
to () and EM
()
(I, ) K
or(+)
for I K
or(+)
and t a
t
embeds I into
EM
()
(I, ) (so t P
I
a
t
P
EM(I,)
hence we may identify t I with

a
t
EM(I, )).
2) I K
or(+)
is strongly
0
-homogeneous when: if n < , s
0
<
I
. . . <
I
s
n1
, t
0
<
I
. . . <
I
t
n1
and s
P
I
1
t
P
I
1
for < n then there is an automorphism h of I
(so mapping P
I
1
onto P
I
1
hence it maps P
I
2
onto P
I
2
) satisfying h(s
) = t
for < n.
3) If
or(+)
[K] and I
0
, I
1
I K
or(+)
and h is an isomorphism from I
0
onto
I
1
then

h is an isomorphism from EM(I
0
, ) onto EM(I
1
, ).
8.11 Observation. 1) I K
or(+)
is strongly
0
-homogeneous i the linear orders
(P
I
1
, <
I
) and (P
I
2
, <
I
) are strongly
0
-homogeneous.
2) There is
lin(+)
0
such that EM
()
(I, ) is strongly
0
-homogeneous for
every I K
or(+)
and EM
()
(I, ) =
2
=1
EM
()
(I P
I
, ).
Proof. Easy, e.g.
2) Let ()() = F
n
: n < , F
n
a (2n+1)-place function and we demand that
in any M = EM(I, ) we have:
(a) P
1
(x
) P
2
(x
m
) F
n
(x
0
, . . . , x
2n
) = x
0
,
(b)
2n
i=0
P
(x
i
) P
(F
n
(x
0
, . . . , x
2n
))
(c) if a
1
<
M
. . . <
M
a
n
and b
1
<
M
. . . <
M
b
n
and a
1
, . . . , a
n
, b
1
, . . . , b
n

P
M
then x F
n
(x, a
j
, . . . , a
n
, b
1
, . . . , b
n
) is an automorphism of (P
M
, <
M
)
mapping a
m
to b
m
for m = 1, . . . , n.
8.11
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

50 SAHARON SHELAH
8.12 Claim. M is saturated if: K is categorical in and
(a) M = EM
(K)
(I, )
(b)
or(+)
[K]
(c) I K
or(+)
satises LS(K) < [I[ cf()
(d) for any < [I[ in I there is a monotonic sequence of length
+
.
Proof. As in [Sh 394, 6] but we prove in details. To prove that M is saturated, let
N
K
M be of cardinality < [I[ and p S
K
(N); then there is J I of cardinality
< [I[ such that N EM
(K)
(J, ), so without loss of generality equality holds. Let
the cardinality of J be so < [I[ hence there is a monotonic sequence t
i
: i <
+
);
without loss of generality it is increasing. So there is an ordinal i() <
+
such that
the interval [t
i()
, t
j
)
I
is disjoint to J whenever i() < j <
and clearly for some

j this interval is innite.
Let I
be like I when we add a copy of just above t

i()
. Let M
= EM
(K)
(I
, )
so M
s
M
and the latter is [I[-saturated ( as LS(K) < [I[ cf() and we know
then M K
M is cf()-saturated by [Sh 394, 6.7=6.4tex]) hence p is realized

by some member of M
. By a claim from [Sh 394], every type over N realized in

M
is already realized in M so we are done.

8.12
Below we can manage using only <
,1
,
,2
, see remarks.
8.13 Remark. Clause (d) in 8.12 is not really necessary but not harmful here.
Why not necessary? E.g., let I
= I Q ordered lexicographically. Now

() if J I
has cardinality < [I[ then for some t
I we have t
Q is
disjoint to J, hence we can proceed as above.
Now given we can nd
such that EM
(K)
(I,
) is isomorphic to EM
(K)
(I
Q, ).
8.14 Denition. We dene partial orders
,1
,
,2
and
,3
on
or(+)
(for
LS(K)) as follows:
1)
1

,

2
if:
(a) (
1
) (
2
) and
(b) EM
(K)
(I,
1
)
K
EM
(K)
(I,
2
) and
(c) EM(I,
1
) = EM
(
1
)
(I,
1
) EM
(
1
)
(I,
2
)
(d) if = 2, 3 then EM(I P
I
1
,
1
) = EM
(
1
)
(I P
I
1
,
2
) for any I K
or(+)
(e) if = 3 then any
1
-automorphism scheme t
1
there is a
2
-automorphism
scheme t
2
which extends it, see denition below.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

2) We say that t is a -automorphism scheme when it is a -automorphism over
P
1
, (n, n
1
)-scheme for some n
1
n < which means that
(a) t is a set of tuples of the formm, m
1
, u,
1
(x
0
, . . . , x
m1
),
2
(x
0
, . . . , x
m1
))
such that m
1
m < , u m, [u[ = n, [u m
1
[ = n
1
and
1
,
2
are ()-
terms
(b) for every m, m
1
, u and
1
(x
0
, . . . , x
m1
) as above for some
2
(x
0
, . . . , x
m1
)
as above the tuple m, m
1
, u,
1
(x
0
, . . . , x
m1
),
2
(x
0
, . . . , x
m1
)) belongs to
t
(c) if (x
0
, . . . , x
m
1
1
) is a ()-term, m
1
m so x
m
1
, . . . , x
m1
are dummy
variables when we use below (x
0
, . . . , x
m1
), and u m, [u[ = n, [um
1
[ =
n
1
then m, m
1
, u, , ) = m, m
1
, u, (x
0
, . . . , x
m1
), (x
0
, . . . , x
m1
)) be-
longs to t
(d) for every I K
or(+)
and t
0
<
I
. . . <
I
t
n1
satisfying t
P
I
1
<
n
1
the set of pairs f
t
,I
[t
0
, . . . , t
n1
] dened below is an automorphism of
EM
()
(I, ).
2A) If we omit over P
1
we omit clause (c).
3) For , I, t, n
1
and t
0
<
I
<
I
t
n1
as above f
t
,I
[t
0
, . . . , t
n1
] is (where M =
EM(I, )) the set of pairs
_
(
M
1
(a
s
0
, . . . , a
s
m1
),
M
2
(a
s
0
, . . . , a
s
m1
)) : there is m, m
1
, u,
1
(x
0
, . . . , x
m1
),
2
(x
0
, . . . , x
m1
)) t, and
s
0
<
I
<
I
s
m1
such that s
P
I
1
< m
1
and u s
= t
|u|
_
.
4) Assume that for m = 1, 2,
m
satises clauses (a),(b),(d) of part (1) and t
m
is
an
m
-automorphism scheme for m = 1, 2. We say that t
2
extend t
1
when t
1
t
2
.
Again for = LS(K) we may drop the .
8.15 Claim. 1) (
or(+)
,
,
) are partial orders for = 1, 2, 3.

2) If
i
: i < ) is a
,
-increasing sequence, <

+
, then it has a <
,
-l.u.b.
; EM(I, ) =
_
i<
EM(I,
i
).
3) Assume that
(a) f is an automorphism of EM
(K)
(I, )
(b) P
I
1
, P
I
2
are dense (in particular with neither rst nor last element)
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

52 SAHARON SHELAH
(c) f commutes with every partial automorphism

h where h is a (nite) partial
automorphism of I extending the identity on t
0
, . . . , t
n1
where t
P
I
1

< n
1
(d) if J K
or(+)
is nite and include t
0
, . . . , t
n1
then f maps EM(J, )
onto itself.
Then for some -automorphism (n, n
1
)-scheme t we have f = f
t
,I
[t
0
, . . . , t
n1
].
4) If in (3) we can add some -automorphism over P
1
when we add the assump-
tion
(e) f is the identity on EM(I P
I
1
, ).
5) In clause (d) of Denition 8.14(2) it is enough that for every i < there
is such I with every E
I
{t
0
,...,t
n1
}
-equivalence class having k members where
s
1
E
I
{t
0
,...,t
n1
}
s
2
i s
1
, s
2
It
0
, . . . , t
n1
and s
1
P
I
1
s
2
P
I
1
, < n
s
1
<
I
t
s
2
<
t
2
I
.
Proof. Easy. E.g., 3),4) Let t = x : x has the form (n, n
1
, u,
2
(x
0
, . . . , x
m1
),
2
(x
0
, . . . , x
m1
), n
1

n, u n,
1
,
2
are () terms and there are s
i
0
<
I
. . . <
I
s
i
m1
for i = 1, 2 such that
ui 1, 2 s
i
= t
(u)
and f(a
1
) = a
2
when we let a
i
=
i
(a
s
i
0
, . . . , a
s
i
n1
).
It is enough to check the clauses (a)-(d) of Denition 8.14(2) and the equality in
the end of the conclusion of part (3).
Clause (a): By inspection of t is a set of tuples of the right form.
Clause (b): By clause (d) of the assumption.
Clause (c): (For part (2)) by assumtion (e).
Clause (d): By part (5) of the claim it suces to prove this for our present I. But
then this is the equality we have promised and proved below.
The equalities: f = f
t
,I
[t
0
, . . . , t
n1
].
The inclusion : Assume f(a
1
) = a
2
so for some nite J I we have a, b
EM(J, ) and without loss of generality t
0
, . . . , t
n1
J. Let s
0
<
I
. . . <
I
t
m1
list J hence tere are ()-terms
i
(x
0
, . . . , x
m1
) such that a
i
=
i
(a
s
m1
) for i =
1, 2. Let m
1
be such that < m
1
s
P
I
1
, let u = < m : s
t
0
, . . . , t
n1
, so
clearly (m, m
1
, u,
1
,
2
) t hence (a
1
, a
i
) = (
1
(a
s
0
, . . . , a
s
m1
),
2
(a
s
0
, . . . , a
s
m1
))
f
t
,I
[t
0
, . . . , t
n1
], so we have proved the inclusion .
The inclusion : If (a
1
, a
2
) f
t
,I
[t
0
, . . . , t
n1
] then there are x = (m, m
1
, u,
1
,
2
)
t and s
1
0
<
I
. . . <
I
s
1
m1
which witnesses it, so < m < m
1
s
1
P
I
1
and
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

a
i
=
i
(a
s
1
0
, . . . , a
s
1
m1
) for i = 1, 2 and u s
1
= t
(u)
. But why x t? There
should be witnesses s
2
0
<
I
. . . <
I
s
2
m1
for this hence < m [ < m
1
s
2
P
I
1
]
and u s
2
= t
(u)
. Let h = (s
1
, s
2
) : < m so h is a partial automorphism

of I which is the identity on t
0
, . . . , t
n1
. By clause (c) of the assumption we are
done.
8.16 Claim. 1) For LS(K) we have
or(+)
[K] ,= .
2) If N is a model, b
n
N for = 1, 2 and n < such that b
n
: n < ) is an
indiscernible sequence over b
3
n
: n < then we can nd proper for K
or(+)
such that () =
N
and
if N
= EM(I, ), s
0
<
I
. . . <
I
s
m1
, t
0
<
I
. . . <
I
t
n1
and s
P
I
1
, t

P
I
2
then the quantier-type which a
s
0
, . . . , a
s
m1
, a
t
0
, . . . , a
t
n1
) realizes in
N
is equal to the quantier free type which b

1
0
, . . . , b
1
m1
, b
2
0
, . . . , b
2
n1
) re-
alizes in N.
3) If in addition (K)
N
and for n
1
n
2
< , m
1
m
2
< we have (N
(K)) c
N
b
1
0
, . . . , b
1
m
1
1
, b
2
0
, . . . , b
2
n
1
1
is a
K
-submodel of N (K) (or just of
(N (K)) c
N
(b
1
0
, . . . , b
1
m
2
1
, b
2
0
, . . . , b
2
n
2
1
)) then
or(+)
[K].
4) Assume
1

,1
2
.
Then we can nd
2

or(+)
[K] such that

()
1

,2

2
() if I K
or(+)
satises P
I
2
,= then EM
(
1
)
(I,
2
) = EM
(
1
)
(I,
2
)
() if I K
or(+)
, then EM
(
1
)
(I P
I
1
,
2
) = EM(I P
I
1
,
1
) actually this
follows from clause ()
() if t
1
is a
1
-automorphism over P
1
, (n, n
1
)-scheme and t
2
is a
2
-automorphism
(n, n
1
)-scheme extending t
1
then there is t
2
a
2
1
, (n, n
1
)-
scheme extending t
1
.
Proof. 1) Because
or
[K] ,= by [Sh 394].

2) Think.
3) The main possibility implies the or just of ... by K being an a.e.c. The statement
itself is easy to check (as we can use just nite I J and then by the axioms of
a.e.c. the case of n
1
, m
2
, is enough).
4) Let the vocabulary
2
(intended to be (
2
)) have the same predicates and
function symbols as (
2
) except that for any function symbol F (
2
)(
1
)
we change its arity: arity
2
(F) = arity
(
2
)
(F) + 1. For I K
or(+)
we dene
M
2
= EM(I,
2
) as follows: let M
2
= EM(I,
2
), now M
2
, a
2
-model is the same
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

54 SAHARON SHELAH
as M
2
except that if F (
2
)(
1
), n = arity
(
2
)
(F) and a
0
, . . . , a
n
M
2
we
dene F
M
2
(a
0
, . . . , a
n
) as follows:
if a
0
, . . . , a
n
EM(I P
I
1
,
1
) then F
M
2
(a
0
, . . . , a
n
) = a
0
and if otherwise then F
M
2
(a
0
, . . . , a
n
) = F
M
2
(a
0
, . . . , a
n1
).
Now it is easy to check.
8.16
For the version (using <
,
, = 1, 2 only) we need
8.17 Denition. 1) We say that t is a weak -automorphism (n, n
1
)-scheme over
P
1
when:
(a), (b), (c) as in Denition 8.14(2) above
(d) for every I K
or(+)
and t
0
<
I
. . . <
I
t
n1
satisfying t
P
I
1
< n
1
the
set f = f
t
,I
[t
0
, . . . , t
n1
] satises
() f is a one to one function
() M
1
f
:= (EM
(K)
(I, )) Dom(f) is a
K
-submodel of EM
(K)
(I, )
() M
2
f
:= (EM
(K)
(I, )) Range(f) is a
K
-submodel of EM
(K)
(I, )
() f is an isomorphism from M
1
f
onto M
2
f
.
2) We say that x is a -task if x has the form (n, n
1
,
1
(x
0
, . . . , x
n1
),
2
(x
0
, . . . , x
n1
)
where
1
,
2
are ()-terms.
3) We say that the [weak] -automorphism over P
1
, (n, n
1
)-scheme t solves the
-task x = (n
x
, n
x
1
,
x
1
(x
0
, . . . , x
n1
),
2
(x
0
, . . . , x
n1
)) when (n, n
1
) = (n
x
, n
x
1
) and
the tuple (n
x
, n
x
1
, n
x
,
x
1
(x
0
, . . . , x
n1
),
x
2
(x
0
, . . . , x
n1
)) belongs to t.
4) We say that the -task x is [weakly] solvable or -solvable if some [weak] -
automorphism (n, n
1
)-scheme solves it.
8.18 Observation. 0) A -automorphism over P
1
, (n, n
1
)-scheme is a weak -
automorphism over P
1
, (n, n
1
)-scheme.
1) If
1

,1

2
and x is a
1
-task then x is a
2
-task.
2) If
1

,2

2
and x is a weakly
1
-solvable
1
-task then x is a weakly
2
-solvable
2
-task.
3) If a -task is solvable then it is weakly solvable.
8.19 Remark. For the weak version, in 8.20 it suces if we weaken the conclusion
()
1

,2

2
() the
1
-task x is
2
-solvable.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

[This simplies the proof.]
8.20 Main Claim. Assume
(a)
1

or(+)
(b) n
1
n < and
1
(. . . , x
, . . . )
<n
,
2
(. . . , x
, . . . )
<n
are (
1
)-terms; for
convenience assume n
1
< n. Let x = (n, n
1
,
2
(. . . , x
, . . . )
<n
,
2
(x
0
, . . . , x
n1
)
<n
)
(c) for every < (2
)
+
there are I and t
0
, . . . , t
n1
such that:
() I K
or(+)
is strongly
0
-homogeneous
() [P
I
1
[
() [P
I
2
[
([P
I
1
[)
() t
0
<
I
. . . <
I
t
n1
so are from I
() t
P
I
1
< n
1
() there is an automorphism of EM
(K)
(I,
1
) which is the identity
on EM
(K)
(I P
I
1
,
1
) and maps
1
(. . . , a
t
, . . . )
<n
to
2
(. . . , a
t
, . . . )
<n
.
Then there are is
2
such that
()
1

,3
2
() there is t such that:
if I K
or(+)
, M = EM
(K)
(I,
1
)
t
0
<
I
. . . <
I
t
n1
and t
P
I
1
< n
1
then there is f such that
(i) f is an automorphism of M which is the identity on EM
(K)
(I
P
I
1
,
1
)
(ii) f(
1
(. . . , a
t
, . . . )) =
2
(. . . , a
t
, . . . )
(iii) f is f
t
2
,I
[t
0
, . . . , t
n1
] and t is a
2
1
, (n, n
1
)-
scheme t.
Remark. 1) Note that EM
(K)
(I P
I
1
,
2
) is equal to EM
(K)
(I P
I
1
,
1
).
2) In the clause (c) of the assumption, there are t
0
<
I
such that is equivalent to
for every t
0
<
I
. . . such that.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

56 SAHARON SHELAH
Proof. For each < (2
)
+
we choose I
, t
: < n) and f
exemplifying clause
(c) of the assumption. We expand M
= EM(I
,
1
) to a model M
+
as follows:
1
(a) Q
M
+
= a
t
: t P
I
for = 1, 2 and Q
M
+
= Q
M
+
1
Q
M
+
2
and
Q
M
+
3
= EM(I
P
I
1
,
0
)
(b) the relation <
M
+
chosen as (a
s
, a
t
) : s <
I
t
(c) H
M
+
m
, (2m+ 1)-place functions (for m < ) are chosen in
0
below
together witnessing I
is strongly
0
-homogeneous
respecting inverses, that is H
m
(H
m
(x, s,
t),
t, s) = x
(d) the function F
M
+
chosen in
1
below
(e) individual constants c
M
+
= a
t
for < n
(f) the predicates P
M
+
: a (
1
)-term) and functions G
M
+
( < )
as in
2
below
(g) Skolem functions (see
3
below).
Now
0
for m < let
(i) h
m
be a ( m + 1)-place function from I
to I
such that: if s
0
<
I
. . . <
I
s
m1
and t
0
<
I
. . . <
I
t
m1
and s
P
I
1
t
P
I
1
then the function x h
m
(x, s
0
, . . . , s
m1
, t
0
, . . . , t
m1
) is an automor-
phism of I
mapping t
to s
for < n; we can add: if t

0
, . . . , t
m1

P
I
then x / P
I
x = h
m
(x, s
0
, . . . , s
m1
, t
0
, . . . , t
m1
)
(ii) H
M
+
m
is the (2m+ 1)-place function from M
to M
dened by:
if s
0
<
I
. . . <
I
s
m1
, t
0
<
I
. . . <
I
t
m1
, s
P
I
1
t
P
I
1
and M
[= a = (a
r
0
, . . . , a
r
k1
) and
r
= h
m
(r
, s
0
, . . . , s
m1
, t
0
, . . . , t
m1
), then
H
M
+
m
(a, a
s
0
, . . . , a
s
m1
, a
t
0
, . . . , a
t
m1
) = (a
r
0
, . . . , a
r
k1
),
in other cases H
M
+
m
(a, b
0
, . . . ) = a
1
recall that f
Aut(EM
(K)
(I
,
1
)) is as in () of clause (c) of the as-
sumption of the claim. Let F
M
+
be a unary function, F
M
+
(b) = f
(b) for
every b EM
(K)
(I
,
1
).
2
For every a M
there are n, t
0
<
I
. . . <
I
t
n1
and (
1
)-term(x
0
, . . . , x
n1
)
such that M
[= a = (a
t
0
, . . . , a
t
n1
). Let n : M
be such that n(a)

is the minimal n for which there are t
0
<
I
. . . <
I
t
n1
and as above. Let
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

<
,n
be a well ordering of the set of (
1
)-terms of the form (x
0
, . . . , x
n1
).
For a M
let
a
be the <
,n(a)
-minimal (
1
)-term (x
0
, . . . , x
n1
) such
that for some t
0
<
I
. . . <
I
t
n1
we have M
[= a = (a
t
0
, . . . , a
t
n1
).
Dene G
M
+
: M
Q
M
3
for < such that if a M
, n = n(a), =
a
then G
M
+
0
(a) <
M
+
. . . <
M
+
G
M
+
n1
(a) so they all belong to a
t
: t I
and M
[= a =
a
(G
M
+
0
(a), . . . , G
M
+
n(a)1
(a)). Let G
M
+
(a) = a when
[n(a), ). Now lastly, let P
M
+
= a M :
a
= .
3
we further add Skolem functions in particular we have A M
(M

(K)) c
M
+
(A)
K
M
(K).
The model we get we call M
+
and without loss of generality (M

+
) does not de-

pend on . Now use a variant of the a.e.c. omitting types theorem, 8.6. So there
are () < (2
)
+
and a model N
+
and b
m
: m < ) in it for = 1, 2 such that
(a) b
m
Q
N
+
, b
2
m
: m < ) is indiscernible over Q
N
+
3
which include b
1
m
: m <
and
(b) b
1
m
: m < ) is indiscernible over b
2
m
: m < and
(c) Th
L
(N
+
) = Th
L
(M
+
()
) (recalling that M
+
()
has Skolem functions),
(d) N
+
omits all quantier free types which M
+
()
omits and
(e) for every m < for arbitrarily large < (2
)
+
,
() Th
L
(M
+
) = Th
L
(N
+
) and
() N
+
omits all the quantier free types which M
+
omits
(f) for some s
0
<
I
. . . <
I
m1
from P
I
for = 1, 2 the quantier free type

of b
1
0
, . . . , b
1
m1
, b
2
0
, . . . , b
2
m1
) in N
+
is equal to the quantier free type of
a
s
1
0
, . . . , a
s
1
m1
, a
s
2
0
, . . . , a
s
2
m1
) in M
+
.
Now there is

or(+)
such that
2
for any I
K
or(+)
, EM(I
) is a (N
+
)-model generated by a
s
: s
I
satisfying s
0
< . . . < s
m1
P
I
1
, t
0
< . . . < t
k1
P
I
2
the
quantier free type of a
s
0
, . . . , a
s
m1
, a
t
0
, . . . , a
t
n1
) in EM(I
) is equal
to the quantier free type of b
1
i
: i < m)b
2
i
: i < k) in N
+
.
[Why? There is
proper for K
or(+)
by 8.16(2). By the choice of N

+
, b
m
: m <
, = 1, 2) and 8.16(3) we know that

or(+)
[K].]
()
0
EM(I,
) N
+
for I K
or(+)
.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

58 SAHARON SHELAH
[Why? As M
+
()
has Skolem functions.]
Now
3

1
<
,1
.
Why? (We prove more). Let I
K
or(+)
.
Assume N
= EM(I
) and let J[N
] K
or(+)
be dened as follows: it is the
set a : N
[= Q(a) linearly ordered by <

N
and let P
J[N
= Q
N
+
for = 1, 2.
So identifying t I
with a
t
we have I
J[N
].
As N
N
+
clearly
()
1
if k m and N
[= a
0
<
a
1
<
. . . <
a
m1
and Q
1
(a
) for < k, Q
2
(a
)
for [k, m) then the L(
1
)-quantier free type which a
0
, . . . , a
m1
)
realizes in N
is equal to the L(
1
)-type which b
1
0
, . . . , b
1
k1
, b
2
k
, . . . , b
2
m1
)
realizes in N
+
; this type is determined by
.
We dene an embedding j of EM(J[N
],
1
) into N
(
1
) as follows for a
0
<
J[N
]
. . . <
J[N
]
a
m1
and (
1
)-term (x
0
, . . . , x
m1
), we decide: (a
0
, . . . , a
m1
) as
interpreted in EM(J[N
],
1
) is mapped to (a
0
, . . . , a
m1
) as interpreted in N
by ()
1
this is an embedding (for
1
). This embedding is onto as
()
2
every c N
is in the closure of a
t
: t J[N
] under the (
1
)-functions
(as interpreted in N
)
which holds as
()
3
M
+
()
so N
+
hence N
omit the type

p(x) =
_
(y
0
, . . . , y
k1
)(
<n
Q(y
) & x = (y
0
, . . . , y
k1
)) : (x
0
, . . . , x
k1
)
1
)
_
.
Also
()
4
j is an isomorphism from EM(J[N
],
1
) onto EM
(
1
)
(I
) mapping a
t
to a
t
for t I
.
Now by the choice of the M
+
s it follows that
()
5
EM(I
,
1
) EM
(
1
)
(I
) and EM
(K)
(I
,
1
)
K
EM
(K)
(I
) for
every I K
or(+)
.
Now
3
follows.
Note that for some proper for K
or(+)
, [
[ we have J[N
] = EM
()
(I
, )
K
or(+)
recalling () = <, P
1
, P
2
and [()[ .
Next there is
2
such that
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

4

1

,2

2
and for every I K
or(+)
if P
I
2
,= then EM
(
1
)
(I,
2
) =
EM
(
1
)
(I,
).
[Why? By 8.16(4).]
Lastly, we have to prove , that is the conclusion of 8.20.
CLAUSE () of the conclusion of the claim:
WHY? By
4
we have just to lift a
1
1
, (n, n
1
)-scheme t
1
to a
2
1
, (n, n
1
)-scheme.
So let I
K
or(+)
be dense and t
0
<
I
. . . <
I
t
n1
, t
P
I
1
< n
1
. So
f
1
= f
t
1
1
,I
[t
0
, . . . , t
n1
] is a well dened an automorphism of EM
(K)
(I
,
1
) over
EM
(K)
(I
P
I
1
,
1
). As in the proof of
3
let N
= EM(I
) and let J = J[N
]
be as there. So I
J hence
() EM(I
,
1
) EM(J,
1
) = EM
(
1
)
(I
) = EM
(
1
)
(I
,
2
).
Let f
2
be f
t
1
1
,J
[t
0
, . . . , t
n1
], it is an automorphism of EM
(K)
(J,
1
) extending f
1
hence it is an automorphism of EM
(K)
(I
,
2
) extending f
1
and it is the identity
of EM
(K)
(J P
J
1
,
1
) which is equal to EM
(K)
(J
p
I
,
2
).
Now assume s
0
<
I
. . . <
I
s
m1
, s
m1
/ p
I
1
and = (x
0
, . . . , x
m1
) is a
(
)-term and t
0
, . . . , t
n1
s
0
, . . . , s
m1
. So we can nd k < and r
0
<
J
. . . <
J
r
k1
and (
2
)-terms
(x
0
, . . . , x
m1
) such that r
(a
s
0
, . . . , a
s
m1
) for
< k and (
1
)-term
(x
0
, . . . , x
m1
) such that (a
s
0
, . . . , a
s
m1
) =
(a
s
0
, . . . , a
s
k1
).
Without loss of generality t
0
, . . . , t
n1
r
0
, . . . , r
k1
and let u = <
k : r
t
0
, . . . , t
n1
, so for each < k for some
(x
0
, . . . , x
k1
) we have
(n, n
1
, u, (x
0
, . . . , x
k1
),
(x
0
, . . . , x
k1
). So let f
2
(
(a
s
0
, . . . , a
s
m1
)) = f
2
(
(a
r
0
, . . . , a
r
k1
)) =
f
2
(
(
0
(a
s
0
, . . . , a
s
m1
), . . . ,
k1
(a
s
0
, . . . , a
s
m1
))).
So by 8.15(3), clearly there is t
2
as required except that we should replace Phi
by
i
but as a
s
m1
/ EM(I
P
I
1
,
2
), there is no problem to correct this.
CLAUSE () of the conclusion:
So assume that I
K
or(+)
and I
[= t
0
< . . . < t
n1
and t
P
I
1
< n
1
and let N
= EM(I
), N
2
= EM(I,
2
).
As I
J[N
], we have J[N
] [= t
0
< . . . < t
n1
and t
P
J[N
]
1
< n
1
. If
n
1
= n the identity can serve as the automorphism, so without loss of generality n
1
<
n hence P
I
2
,= . Let t
be c
N
for = 0, . . . , n1, recalling clause (e) of

1
. By the
choice of the functions H
M
+
n
(see clause (c) of
1
above) there is an automorphism
h of the linear order J[N
] such that
5
h
1
(t
) = t
Clearly
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

60 SAHARON SHELAH
6
h induces an automorphism

h of the model EM
(
1
)
(I
) which is equal to
EM(J[N
],
1
) by

h((a
s
0
, . . . , a
s
k1
) = (a
h(s
0
)
, . . . , a
h(s
k1
)
) when s
0
, . . . , s
k1

J[N
] and is a term in (
1
).
So clearly
7
h
1
(t
0
), . . . , h
1
(t
n1
) are as in clause (c) of the assumption of 8.20 for the
linear order J[N
] K
or(+)
.
Now the property
() which is inherited by N
as N
+
N
M
+
()
(and the
choice of F
M
+
()
in
1
above and the choices of t
(and of the c
M
+
above) gives
for some automorphism f of EM

(K)
(J[N
],
1
) we have
() f EM
(K)
(J[N
] Q
N
1
,
1
) is the identity
() f(
1
(. . . , a
h
1
(t
)
, . . . )
<n
) =
2
(. . . , a
h
1
(t
)
, . . . )
<n
because
f(
1
(. . . , a
t
, . . . )
<n
) =
2
(. . . , a
t
, . . . )
<n
() f(b) = F(b) that is f(b) = F
EM(I
)
(b) for every b EM(I
);
see the choice of F
M
+
()
above.
Hence also f
=

hf
h
1
is an automorphism of N
(K) = EM
(K)
(J[N
],
1
) =
EM
(K)
(I
).
Let us check the demands listed in () of 8.20. First half of Subclause ()(i)
holds by
, and Subclause ()(ii) there holds by clause () of
above. Next
Subclause ()(iii) there holds for
by our choice of f and F

M
+
()
, i.e., clause ()
of
and by h being denable by the H

n
s and M
+
having Skolem functions.

More fully, let

t = t
: < n),
= t
: < n), a
1
=
1
(a
t
0
, . . . , a
t
n1
). For
every b N
clearly for some = (x

0
, . . . , x
k1
), a (
1
)-term and s
<
I
. . . <
J
s
k1
we have b = (a
s
0
, . . . , a
s
k1
) where b P
N
and s
= G
N
(b), so
7
by
6
b
1
=:

h
1
(b) = H
N
n
(b, a
t
0
, . . . , a
t
0
, . . . )
= H
N
n
(b, G
N
0
(a
1
), G
N
1
(a
1
), . . . , G
N
n1
(a
1
), c
0
, . . . , c
n1
)
b
2
=: f(b
1
) = F
N
(b
1
)
7
we ignore the case that in
1
(x
1
, . . . , x
n1
) some of the variables are dummy variables just
use a
t
instead G
N
1
(a
1
).
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

and recalling the end of clause
6
we have
b
3
=:

h(b
2
) = H
N
n
(b
2
, a
t
0
, . . . , a
t
n1
, a
t
0
, . . . , a
t
n1
)
= H
N
n
(b
2
, c
0
, . . . , c
n1
, G
N
0
(a
1
), . . . , G
N
n1
(a
1
)).
Composing clearly f
(b) = (
hf
h
1
)(b) =

h(f(
h
1
(b)) =

h(f(b
1
)) =

h(b
2
) = b
3
is
equal to
(b) for the suitable term
(x) = (x, G
0
(a
1
), . . . , G
n1
(a
1
), c
0
, . . . , c
n1
)
of (
) which does not depend on b. As M
()
has Skolem function we can re-
place
(x) by F
(x, y) for some function symbol F
). But we need f
is
the identity on EM(P
I
1
,
1
) which is for the second half of Subclause ()(i); let
b EM(I
P
I
1
,
1
) or just b EM(P
J[N
]
1
,
1
); so for some term of (
1
)
and s
0
, . . . , s
m1
P
J[N
]
1
= Q
N
1
we have N
[= b = (a
s
0
, . . . , a
s
m1
), hence
(as N
N
+
and the choice of F
M
+
)
8
f
(b) = (
hf
h
1
)((a
s
0
, . . . , a
s
m1
)) =
h(f(
h
1
((a
s
0
, . . . , a
s
m1
)))) =
(
h(f((a
h
1
(s
0
)
, . . . , a
h
1
(s
m1
)
)))).
[The last equality by the denition of

h from h, see
6
.]
But we are assuming that s
0
, . . . , s
m1
Q
[N
]
1
and by the choice of h in
5
we
conclude that h
1
(s
0
), . . . , h
1
(s
m1
) Q
N
1
hence by clause () of
above we
have
9
f((a
h
1
(s
0
)
, . . . , a
h
1
(s
m1
)
)) = ((a
h
1
(s
0
)
, . . . , a
h
1
(s
m1
)
)).
By
8
+
9
f
(b) =

h((a
h
1
(s
0
)
, . . . , a
h
1
(s
m1
)
) = ( a
hh
1
(s
0
)
, . . . , a
hh
1
(s
m1
)
)
= ( a
s
0
, . . . , a
s
m1
) = b.
As b was any member of EM
(K)
(I
P
I
1
,
1
) we are done proving the second half
of subclause ()(i) of .
We have shown above strongly version of denability for
, so by the way
2
was constructed from
it follows that also subclause ()(iii) of holds that is

8.16(4)().
8.20
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

62 SAHARON SHELAH
8.21 Claim. Assume
(a) K is an a.e.c. with amalgamation, categorical in
(b) the M K
is
+
-saturated (holds if cf() > )
(c) LS(K).
Then every M K of cardinality
(2
)
+ (or just
()
if 2
LS(K)
) is
+
-saturated.
Proof. If M is a counterexample, let N
K
M, |N| and p S(N) be omitted
by N. By the omitting type theorem for abstract elementary classes (see 8.6, i.e.
[Sh 88]), we get M
, N
K
M
, M
omitting p a contradiction.
1.9
8.22 Claim. Assume
(a) LS(K)
(b) for every < (2
)
+
there are M
<
K
N
(so M
,= N
), |M
and
p S(M
) such that c N
pE
tp(c, M
, C).
1) For every > there are M <
K
N in K
and p S(M
) as in clause (b).
2) Moreover, if is proper for orders as usual, [()[ ,
(2
)
+ , K categorical
in and I a linear order of cardinality , then we can demand M = EM
(K)
(I, ).
Proof. Straight.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

8.23 Conclusion. 1) For LS(K) and
or(+)
there are () < (2
)
+
and
such that
(a)
,3
so

or(+)
(b) if x = n, n
1
,
2
(. . . , x
m
, . . . )
m<n
,
2
(x, . . . , x
m
, . . . )
m<n
) and
2
and t
:
< n) satises
()
of clause (c) of 8.20 holds with (x,
) here standing
for (x,
1
) there (so x is a
-task) then () from the conclusion of 8.20

holds.
2) We can replace () < (2
)
+
by () <
; on
see, e.g., [Sh:c, VII,5].

Proof. 1) We iterate 8.20 recalling that
1

,3

2

or(+)
[K] implies that

I K
or(+)
, I = I P
I
1
EM
(
1
)
(I P
I
1
,
2
) = EM(I P
I
1
,
1
). We choose
by induction on such that
: ) is
,3
-increasing continuous,
in
or(+)
. Let
0
= and in limit stages we take unions. In the induction step,
= + 1 is by 8.20, with (
) here standing for (

1
,
2
) there and x = x
as
in clause (b) of the assumption of 8.20 is chosen such that for every < and any
-tasks x (i.e., as in clause (b) for
) for some < , x

B
= x, this is done by
bookkeeping.
2) Reect.
8.23
8.24 Denition. 1) For

or(+)
[K] let (
) be the minimal ordinal ()

such that
()
> LS(K) and if x is a
-task (see Denition 8.17(2)) and
of
clause (c) of the assumption of 8.20 fails for some then it fails for some ()
(hence for = ()).
2) For = 0, 1 let
) =
() =
(
)+(
)
and () =
1
(
).
8.25 Observation. 1) If
1
is from
or(+)
[K] then (
1
) <
< (2
)
+
.
8.26 Remark. 1) Actually because of K has amalgamation, it is easier to prove
the following weaker variant of 8.23 replacing there is an automorphism f of
EM
(K)
(I,
) over EM
(K)
(I P
I
1
,
) which maps b
1
to b
2
by there are models
N
, b
1
1
, b
2
N
and EM
(K)
(I P
I
1
,
)
K
N

K
EM
(K)
(I,
) and iso-
morphism f from N
1
onto N
2
over EM
(K)
(I P
I
1
,
) mapping b
1
to b
2
. This is
actually enough.
2) We can get also somewhat stronger results, see [Sh:F657].
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

64 SAHARON SHELAH
9 Small Pieces are Enough and Categoricity
Using the context below is justied by the previous sections.
9.1 Context. a) K is categorical in .
b)
as in 8.23.
9.2 Main Lemma=Local Lemma. If M
K is a saturated model of cardinality

,
1
() < cf() , see 8.24(2) then S(M
) has character or locality

1
(), (see Denition 1.8, i.e., if p
1
,= p
2
are in S(M
) then for some N

K
M, N K
1
()
we have p
1
N ,= p
2
N).
Proof. So
as in 8.23. Choose I such that (easily exists by 8.11(2))
1
(a) I K
or(+)
is strongly
0
-homogeneous
(b) P
I
1
has cardinality
(c) I has cardinality
(d) in P
there is a monotonic sequence of length

+
whenever
< [P
I
[ for = 1, 2.
Let N
= EM
(K)
(I,
) K
and M = EM
(K)
(I P
I
1
,
) so M K
is
saturated (see 8.12 above) hence without loss of generality M
= M. Similarly in
N
every p S(M
) is realized. Assume toward contradiction that p

1
,= p
2
are
from S(M
) but p
1
/E
1
()
= p
2
/E
1
()
, i.e., M
K
M, M
1
()
p
1
M
=
p
2
M
. We can nd b
1
, b
2
N
which realize p
1
, p
2
respectively. As we can add
dummy variables we can nd n and t
0
<
I
< . . . <
I
t
n1
and terms
1
,
2
of (
)
such that b
(a
t
0
, . . . , a
t
n1
). Let n
1
n be such that t
P
I
1
< n
1
; again
without loss of generality, n
1
< n.
Recall (8.11(1)) that the linear order (P
I
1
,
I
) is strongly
0
-homogeneous. We
can nd a strongly
0
-homogeneous J
1
P
I
1
of cardinality
1
() which includes
t
: < n P
I
1
. Let J
2
P
J
2
be of cardinality such that it contains a
monotonic sequence of length
+
for every < and is strongly
0
-homogeneous
and t
: < n P
I
2
J
2
.
Let J = I (J
1
P
I
2
) and let M
= EM
(K)
(J P
J
1
,
) = EM(J
1
,
)
and N
1
= EM
(K)
(J,
). Easily M
K
N
, b
1
, b
2
N
, M
K
()
, J is
strongly
0
-homogeneous (see 8.10(1)) and by the choice of p
1
, p
2
, b
1
, b
2
we have
tp
K
(b
1
, M
, N
) = tp
K
(b
2
, M
, N
). But N
is saturated by 8.12 hence there is an

automorphism of N
over M
mapping b
1
to b
2
. As
1
() by the choice of
, i.e., by 8.23 we can conclude that there is an automorphism of N
over M
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

mapping b
1
to b
2
, contradiction to p
1
,= p
2
. We have to check
; so note that
[P
J
1
[ = [J
1
[ =
0
(
), P
J
2
= [J
2
[ =
1
(
), etc.
9.2
9.3 Claim. If T is categorical in , LS(K) () < and M
i
: i < ) an
<
K
-increasing sequence of
+
-saturated models then
_
i<
M
i
is
+
-saturated.
Remark. 1) Hence this holds for limit cardinals > LS(K).
2) The addition compared to 6.7 are the cases cf() =
+
,
++
, e.g. =
+
. The
only case needed is =
++
(used in ()
8
of the proof of 9.5).
[Saharon: check again! as
+
= is trivial.]
Proof. Let M
=
_
i<
M
i
and assume M
is not
+
-saturated. So there are N
K
M
of cardinality and p S(N) which is not realized in M
. Choose p
1
S(M
)
such that p
1
N = p.
Without loss of generality N is saturated (by 6.7, or think).
Let = (), without loss of generality = cf() and each M
i
has cardi-
nality
+
hence i < M
i
is saturated.
We claim
there are i() < and N
K
M
i()
of cardinality such that p
1
does not
-split over N
.
Why? Assume toward contradiction that this fails. The proof of
splits to two
cases.
Case I: cf() .
We can choose by induction on i < = cf() models N
0
i
, N
1
i
such that
(a) N
0
i

K
M
i
has cardinality
(b) N
0
i
is increasing continuous
(c) N
0
i
<
1
,
N
0
i+1
(d) N
0
i

K
N
1
i

K
M
(e) N
1
i
has cardinality
(f) p
1
N
1
i
does -split over N
0
i
(g) for , < i, N
1
N
0
i
.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

66 SAHARON SHELAH
There is no problem to carry the denition and then N
1
i

_
j<
N
0
j
and
N
0
i
: i < ), p
1

_
i<
N
0
i
contradicts 6.3.
Case II: cf() > .
Now by 3.3
() for some N
K
M
of cardinality we have p
1
does not -split over N
.
As = cf() > , for some i() < we have N
K
M
i()
. This ends the
proof of
.
So i(), N
are well dened. Without loss of generality N
is saturated. Let c C
realize p
1
. We can nd a
K
-embedding h of EM
(K)
(
+
+
+
,
) into C such that

(a) N
is the h-image of EM
(K)
(,
)
(b) h maps EM
(K)
(
+
,
) onto some M
K
M
i()
(c) every d N and c belong to the range of h.
By renaming, h is the identity, clearly for some <
+
we have
N c EM
(K)
( [
+
,
+
+),
), so some list

b of the members of N is
(. . . , a
i
, . . . , a
+
+j
, . . . )
i<,j<
(assume > for simplicity) and
c =
(. . . , a
i
, . . . , a
+
+j
, . . . )
iu,jw
(u, w
+
nite as, of course, only nitely
many a
i
s are needed for the term
) and without loss of generality u w .

For each <
+
we dene

b
= (. . . , a
i
, . . . , a
(1+)+j
, . . . )
i<,j<
and
c
(. . . , a
i
, . . . , a
(1+)+j
, . . . )
i,j
and stipulate

b
+
=

b, c
+
= c and let q =
tp(
bc, N
, C). Clearly
: <
+
)
bc) is a strictly indiscernible sequence

over N
and M
c, so also
:
+
M
is strictly indiscernible over

N.
[Why? Use I
+
+
+
which is a strongly
++
saturated dense linear order and
use automorphisms of EM(I,
) induced by an automorphism of I.]

As c realizes p
1
clearly tp(c, M
) does not -split over N
hence by 9.2 recalling

that M
i()
is a saturated model of cardinality
+
necessarily tp(
c, N
, C) is
the same for all
+
, hence <
+
tp(
+
, N
, C) = q, so by the
indiscernibility <
+
tp(
, N
, C) = q.
Similarly for some q
1
,
<
+
tp(
, N
, C) = q
1
.
If q ,= q
1
, we get the (, 1, 0)-order property (see Denition 4.3) contradiction to
(or 4.8).
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

Hence necessarily
+
&
+
& ,= tp(
, N
, C) = q. For
=
+
, = 0 we get that c
M
i()

K
M
realizes tp(c
, N, C) = p
1
N as
desired.
9.3
We could have mentioned earlier parts (1) + (2) of the following observation.
9.4 Observation. 1) If M is -saturated, > LS(K) and < and N
K
M, N
K
then there is N
, N
K
M, N
and every p S(N) realized in M is

realized in N
.
2) Assume N
i
: i ) is
K
-increasingly continuous, <
+
is divisible by
, N
i
K
, N
i

K
M, M is -saturated, and every p S(N
i
) realized in M is
realized in N
i+1
then
(a) if = , LS(K) < = cf() , then N
is -saturated
(b) if = , > LS(K), then N
is saturated.
3) In part (1) we can add: N
is saturated and even saturated over N (here we use

9.3).
9.5 Theorem. (The Downward Los theorem for successors).
If is successor
(2
()
)
+ < , then K is categorical in .
9.6 Remark. 0) In fact, we can replace
(2
()
)
+ by (()).
1) We intend to try to prove in future work that also in K
>
we have categoricity
and deal with not successor. This calls for using P
(n)-diagrams as in [Sh 87a],

[Sh 87b], etc.
2) We need [and can have one but not here] some theory of orthogonality and
regular types parallel to [Sh:a, Ch.V] = [Sh:c, Ch.V], as done in [Sh:h] and then
[MaSh 285] (which appeared) and then (without the upward categoricity) [KlSh
362], [Sh 472]. Then the categoricity can be proved as in those papers.
Proof. Let K
= M K : M is ()
+
-saturated hence of cardinality > ().
So
()
0
there is M K
which is -saturated
[why? by 2.6, 1.7, as is regular]
()
1
K
is closed under
K
-increasing unions
[Why? By 9.3 (or 6.7)]
()
2
if
(2
()
)
+ and M K
then M K
, moreover M is
(2
()
)
+-saturated
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

68 SAHARON SHELAH
[Why? Otherwise by 8.6 there is a non ((
))
+
-saturated M K
con-
tradicting ()
0
, or use 8.21. For the Moreover use 8.7 instead of 8.6.
Assume M K
,
(2
()
)
+ and M is not
(2
()
)
+-saturated. Let
N
K
M be of minimal cardinality such that some p S(N) is omitting
so
(2
()
)
+ > |N| > () hence
(2
()
)
+(|N|) =
(2
()
)
+ |M|. By
9.4, without loss of generality N is saturated, hence by 9.2 for every c M,
for some N
c

K
M of cardinality () we have: tp
K
(c, N
c
, M) ,= p N
c
.
Now 8.7 applies and gives contradiction.]
()
3
if M K and p S(M) then for some M
0

K
M, M
0
K
)
and p does
not (
)-split over M
0
[why? by 3.3, 1.7]
()
4
Denition: for [(), ) and M K
and p S(M) we say p is

minimal if:
(a) p is not algebraic which means p is not realized by any c M
(b) if M
K
M
and p
1
, p
2
S(M
) are non-algebraic extending p,

then p
1
= p
2
()
5
Fact: if M K
is saturated, [(), ), then some p S(M) is

minimal
[Why? If not, we choose by induction on for every

2 a triple
(M
, N
, a
) and h
,
for such that:
(a) M
<
K
N
and a
(b) M
: ) is
K
-increasingly continuous
(c) M
<
1
,
M
(+1)
(d) h
,
is a
K
- embedding of N
into N
which is the identity on

M
and maps a
to a
(e) if ,

2, then h
,
= h
,
h
,
(f) M
0
= M
1
but
tp(a
0
, M
0
, N
0
) ,= tp(a
1
, M
1
, N
1
)
(g) M
<
K
C.
No problem to carry the denition and let = Min : 2
> and
choose M <
K
C such that |M| and
>
2 M
M hence

2 M
M so tp(a
, M, C) :

2 is a subset of S(M) of
cardinality 2
> . So we can get a contradiction to stability in , i.e., to

1.7].
()
6
Fix a saturated M
K
(
)
and minimal p
S(M
)
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

()
7
if M
K
M K
<
, then p
has a non-algebraic extension p S(M),

moreover; if M is saturated, it is unique and also p is minimal
[Why? Let M
K
<
be saturated such that M
K
M
, |M
| = |M|;
and if |M| = (), M
is brimmed over M
(see Denition 2.7). Exis-

tence of non-algebraic p
S(M
) holds by as (M
, M) is isomorphic to
(EM
((
),
), EM
(|M
|,
).
Now for (p
, M
) uniqueness modulo E
()
follows from the denition of p
is minimal hence uniqueness for (M
, p
) holds by the locality lemma 9.2.

So we have proved the moreover. When M is not saturated applying
what we have proved to a saturated extension M
of M of cardinality |M|
we get a non-algebraic p
S(M
) extending p, now p
M is as required].
Let
1
be the predecessor of .
()
8
there are no M
1
, M
2
such that:
(a) M
K
M
1

K
M
2
(b) M
1
, M
2
are saturated of cardinality
1
(c) M
1
,= M
2
(d) no c M
2
M
1
realizes p
[Why? If there is such a pair (M

1
, M
2
), we choose by induction on <
, N
1
which is
K
-increasing continuous, each N
is saturated, N
0
=
M
1
, N
,= N
+1
and no c N
+1
N
realizes p
. If we succeed, then
N =
_
<
N
is in K
(as N
,= N
+1
!) but no c NN
0
realizes p
(why? as : c / N
is an initial segment of , non-empty as 0 is in

so it has a last element , so c N
+1
N
so realizes p
, contradiction);
hence N is not saturated, contradiction to categoricity in by ()
0
. For
= 0, N
0
= M
1
is okay by clause (b). If is limit < , let N
=
_
<
N
,
clearly N
1
and it is saturated by 9.3. If = +1, note that as N
, M
1
are saturated and in K
1
and
K
-extends M
which has smaller cardinality,

there is an isomorphism f
from M
1
onto N
which is the identity on M
.
We dene N
such that there is an isomorphism f

+
from M
2
onto N
extending f
. By assumption (b), N
1
is saturated and by assumption
(c), N
,= N
+1
, and by assumption (d), no c N
+1
N
realizes p
(as
f
= the identity). So as said above, we have derived the desired

contradiction].
()
9
if M K
<
and M
K
M <
K
N, M has cardinality
=
(2
()
)
+, then
some c NM realizes p
.
[Why? Assume this fails, hence by ()
2
, M, N are
-saturated. So we
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

70 SAHARON SHELAH
can nd saturated M
K
M, N
K
N of cardinality
such that M
=
N
M, M
,= N
(why? by observation 9.4(1)-(3)). So still no c N
realizes p
. We would like to transfer (using the appropriate omitting type

theorem) this situation from
to
1
; the least trivial point is preserving
the saturation. But this can be expressed as: is isomorphic to EM(I,
)
for some linear order I for appropriate , and this is easily transferred].
()
10
if M K
has cardinality
=
(2
()
)
+ then it is
-saturated
(so K
).
[Why? By ()
2
.]
()
11
if M K
has cardinality
, then M is saturated
[why? Assume not; by ()
10
, M is
-saturated let be such that M is -

saturated but not
+
-saturated; by ()
10
,
, without loss of generality

M
K
M. Let M
0

K
M be such that M
0
K
and some q S(M

0
) is
omitted by M and without loss of generality M
K
M
0
.
Now choose by induction on i <
+
a triple (N
0
i
, N
1
i
, f
i
) such that:
(a) N
0
i

K
N
1
i
belong to K
and are saturated

(b) N
0
i
is
K
(c) N
1
i
is
K
(d) N
0
0
= M
0
and d N
1
0
realizes q
(e) f
i
is a
K
-embedding of N
0
i
into M and f
0
= id
M
0
(f) for each i, for some c
i
N
1
i
N
0
i
we have c
i
N
0
i+1
(g) f
i
is increasing continuous.
If we succeed, let E = <
+
: limit and for every i < and c N
1
i
we have (j <
+
)(c
j
= c) (j < )(c
j
= c). Clearly E is a club of
+
,
and for each E, c
belongs to N
1
=
_
i<
N
1
i
so there is i < such that
c
N
1
i
, so for some j < , c = c
j
so c
= c
j
N
0
j+1

K
N
0
, contradiction
to clause (f).
So we are stuck for some , now ,= 0 trivially. Also not limit by 9.3,
so = + 1. Now if N
0
= N
1
, then f
(d) M realizes q, where d is

from clause (d), a contradiction, so N
0
<
K
N
1
. Also f
(N
0
) <
K
M by
cardinality consideration. Now by ()
9
some c
N
1
N
0
realizes p
.
We can nd N

K
M such that f
(N
0
) <
K
N
, N
saturated (why?
by 9.4(3)).
Again by ()
9
we can nd c
(N
0
) realizing p
. By ()
5,7
, the
uniqueness part clearly tp(c
, f
(N
0
), M) = f
(tp(c
, N
0
, N
1
)) so we can
nd N
1
which is a
K
-extension of N
1
and a
K
-embedding g
of N
into N
1
which extends f
1
and maps c
to c
. Without loss of generality

(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

N
1
is saturated. Let N
0
= g
(N
) and N
1
, c
were already dened. So we

can carry the construction, contradiction, so ()
11
holds].
()
12
K
is categorical in every [
(2
()
)
+, )
[why? by ()
11
every model is saturated and the saturated model is unique].
9.5
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

72 SAHARON SHELAH
Glossary
0 Introduction
Denition 0.1: K is an a.e.c.
Denition 0.2: 1) K
, h a
K
-embedding, K has amalgamation, -amalgamation,
JEP and M <
K
N
Denition 0.3: LS(K), LS
(K)
Claim 0.4: directed unions
Claim 0.5: Representing an a.e.c. by EC(, )
Claim 0.6: Existence of EM models see 8.6, CHECK!
1
Hypothesis 1.1: (a) K a.e.c.
(b) K has amalgamation and JEP
(c) K
,= for every
Convention 1.2: There is monster C
Denition 1.3: 1) K is categorical in
2) I(, K)
Denition 1.4: tp(a, M, N)
2) tp(a, M) = tp( a, M, C) =
a
M
= a/m; dene M is -saturated where > LS(K)
3),4) S
(M), S
(M) = S
1
(M)
5) p
0
= p
1
M
Denition 1.5: K is stable in
Convention 1.6: as in 0.6
Claim 1.7: If K is cateogorical in then K is stable in every , LS(K) < and
any M K
is cf()-saturated [proof] CHECK

Denition 1.8: 1) For LS(K), E
= E
1
[K] = (p
1
, p
2
) : p
1
, p
2
S
(M) and
(N K
)(N
K
M p
1
N = p
2
N)
2) p S
m
(M) is -local if p/E
is a singleton
3) K is -local if every p S
<
(M) is -local
4) c realizes p/E
is in M
if M
K
M
, c M
, tp(c, M, M
)E
p.
Remark 1.9: 1) E is an equivalence relation
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

2) In previous contents E
LS(K)
is equality; a place to encounter some of the
diculty
3) -local -compactness
Claim 1.10: 1) N
0
is
K
-maximal member
2) p
s
M
1
is well dened, unique when M
1

K
M
2
, p
2
S
(M
0
)
3) types have extension
4) (p M
1
) M
0
= p M
0
Claim 1.11: If M
i
(i w) is
K
-increasing continuous, p
n
S
(M
n
), p
n
= p
n+1

M
n
then there s a limit p
w
S(M
w
), n < p
n
= p
M
n
Remark 1.12:
2 Variants of saturated
Denition 2.1: Assume K is stable in , <
+
1) M <
0
,
N
2) M <
1
,
N
3) default for is...
Lemma 2.2: Basic properties; assume K stable in and <
0) when
,
2
|
ell
,
1) For M inK
there are N
s such that N
,
N
2) Monotonicity
3) Preservation under increasing
4),7)
0
implies universality
5),6) Uniqueness for
1
8) M <
1
,
N is cf()-saturated [proof later]
Discussion 2.3: On [Sh 300], [Sh 87a], [Sh 87b]
Remark 2.4: On
1
,
Denition 2.5: Recall -saturated
Proof of 2.2(8):
Claim 2.6: If K is categorical in , M K
, cf() > then M is (,

+
)-model
homogeneous
Denition 2.7: N is (, )-brimmed
Claim 2.8: 1) Restating properties for brimmness
Discussion:
3 Splitting
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

74 SAHARON SHELAH
(7 line intro)
Content 3.1: C
Denition 3.2: S(M) does -split over N
K
M
Claim 3.3: From stability to non-splitting [proof]
Conclusion 3.4: When p is the E
-unique extension of p N
1
which does not -split
over N
0
4 Indiscernibility and E.M. Models
Notation 4.1: A function h : Y M is the same as sequence h(t) : t Y )
Denition 4.2: 1) Let h
i
: Y C for i < i
1) h
i
: i < i
) is an indiscernible sequence of character < (...use automor-

phisms
2) h
i
: i < i
) is an indiscernible set
3) h
i
: i < i
) is strictly indiscernible sequence if.... (there are ...)

4) h
i
: i < i
) has localness
Denition 4.3: 1) K has the (, )-order property
2) K has the 9
1
,
2
, )-order property
Claim 4.5: 1) Any strictly indiscernible sequence (over) A is an indiscernible set
(over A)
2) We can omit strictly , we can add of character <
Claim 4.6: 1) Existence of
or
LS(K)+|Y |
[K] immitating

h = (h
i
: i < i
)) if
i
<
1,1
(LS(K) +[Y [), (or use h
= h
i
: i < for <
1,1
(LS(K) +[Y [)
2) If h
i
: i < i
) is an indiscernible sequence of character

0
, greater similarly
3) Apparent weaking of assumption in 4.3
4) Variants of 4.3(1),4.7(2)
5) Another building of
Lemma 4.7: If there is a strictly indiscernible sequence which is not an indiscernible
set of character
0
with sequence of length then K hsa the ()-order property
Claim 4.8: 1) If K has the -order property then

I(, K) =

I for every

I( >
( + LS(K))
+
(and more)
2) If K has the (
1
,
2
, )-order property and =:
1
+
2
+ then for some
M K
we have [S
2
(M)/E
[ >
Denition 4.9: 1) p S
m
(N) divides or mu-divides over M
K
N (indiscernible
copies of N over M)
2)
(K),
(K) (long diviidng + types for K
), sets of cardinals
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

3)
,
(K) and
M,
(K), similarly with models in K
and -dividing
Remark 4.10:
Fact 4.11: 1) Equivalent variant of 4.9(1)
2) K
(K), = cf() K
(K); similarly for
,
(K)
3) K
(K)
(K), similarly for

,
Denition 4.12: Assume M
K
N, p S(N), M K
, LS(K)
1) p does -strongly split over M
2) p explicitly -strongly splits over M
Claim 4.13: 1) Strongly splitting implies dividing if ()
,
0
,
0
(see below)
Claim 4.14: 1) If ()
,,
fails then K has the (, + LS(K))-order property
2) for + LS(K) then for some M K
u
, S(M) for some <
3) weaker version
4) a variant
Claim 4.15: EM model M, if the skeleton converges in any N, M
K
N then we
get e.g. instability, order property [proof]
Claim 4.16: for K categorical, existence of strictly indiscernible sub-sequences
[proof]
Observation 4.17: indiscernible in

Denition 4.18: M is -strongly saturaetd (also automorphic)

4 (Second version)
Denition 4.2: indiscernible, strictly indiscernible (by EM)
Denition 4.3: (, )-order, g(a
i
) = , [A[ = , (
1
,
2
, )-order
Observation 4.4: monotonicity
Claim 4.5: triviality
Claim 4.6: existence of indiscernibles toward EM
Claim 4.7: indiscernible sequence not set implies order
Claim 4.8: -order I(, K) = 2
Denition 4.9: p-divide, -divide,

,
[K]
Denition 4.12: -split, -strongly split
Claim 4.13: implication
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

76 SAHARON SHELAH
Claim 4.14: getting order property, unstability
Claim 4.15: almost t I, a
t
/
b equal
Claim 4.16: every a
i
: i < ) contains large strictly indiscernible sub-sequence
Denition 4.18: M strongly saturated
5 Rank and superstability
Denition 5.1: R(p)
Denition 5.2: (, 1)-superstable
Claim 5.3: failure from 4.13 (, 1)-superstability fail
Claim 5.4: if K not (, 1)-superstable then
i
: i w + 1), etc
Claim 5.5: in 5.4 get unstable in <
0
Remark 5.6: discussion
Claim 5.7: categoricity (, 1)-superstable,
(K) =
Claim 5.8: long splitting not saturated
Claim 5.9: from Ramsey cardinal....getting I(, K) large
Claim 5.10: [S(M)/E
M
[, |M|
(2
)
+, LS(K) implies not (, 1)-
superstable and even
(K) large.
6 Existence of many non-splitting
Question 6.1: union of <
1
,
-increasing chains of every type in S(N
) does not split

over some N
i
(and have many extensions)
Observation 6.1: Implications related to 6.1 p ,= q S(N
) (i < )(p i ,=
q i)
Lemma 6.3: yes, by categoricity but <
Theorem 6.5: Assume categoricity in and the M K
is
+
-saturated, LS(K) <
< .
1) M <
1
,
N saturated.
2) There is a saturated M K
.
3) M
1
,
N
1

=
M
N
2
.
Claim 6.7: union of saturated chains is saturated [2003/10/13, changes 6.5-6.7]
Claim 6.8: (K cateogorical in , cf() > LS(K) or just the M K
are
+
-saturated).
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

1) If I J every cut if I realized in JI is realized innitely after (or slightly
closed ) then every p S(EM
(K)
(I, ) is realized in EM
(K)
(J, ), when I
K
lin
.
7 More on splitting
Hypothesis 7.1: increasing union of -saturated is -saturated when (LS(K), )
Conclusion 7.2: M K
saturated, p S(M), then p does not -split over some

M
<
,
M
Fact 7.3: M
is
1
,
-increasing, p S(M
3
) does not split over some M
0
then
R(p) = R(p M
2
)
Claim 7.4: [categoricity] q not stationary q split
Claim 7.5: additivity of non-splitting.
8 Existence of ncie
Context 8.1: K a.e.c. (K
,= )
Remark 8.2: On variants
Denition 8.3:
or
=
or
[K],
or
=
or
[K]
=
LS(K)
Denition 8.4: 1)
, partial order on
or
.
2)
2
is an inessential extension of
1
,
1

ie

2
.
3)
lin
.
4)
, a partial order on
or
.
Claim 8.5: basic properties
Lemma 8.6: The a.e.c. omitting type theories
Lemma 8.7: A two cardinal version [proof]
Remark 8.8: 1) Can use q
1
1
N
n
: n < ).
2) c( a, M).
Denition 8.9: 1) K
or()
: order expanded by P
I
, P
I
1
an initial segment P
I
2
its
compliment.
2)
or()
[K], class of relevant

Denition 8.10: 1)
or()
producing reasonable extension of I K

or()
.
2) I K
or(+)
is strongly
0
-homogenous.
3) for h isomorphic from I
0
to I
1
both I K
or()
,
h is the induced homo-

morphism on the models
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

78 SAHARON SHELAH
Observation 8.11: 1) When I K
or(+)
is strongly
0
-homogeneous.
2) There is
lin(+)
0
guaranteeing this [short proof]
Claim 8.12: M is saturated when K is categorical in , M rm EM(I, ), etc.
[proof]
Remark 8.13: The I wide not necessary
Denition 8.14: 1) We dene
,
partial orders on
or(+)
2) t is a -automorphism scheme over p

1
2A) Without over P
1
3) Dene f
t
,I
[t
0
, . . . , t
n1
] the automorphism
4) We use t
1
t
2
Claim 8.15: 1) Basic properties of the objects from 8.14, [proof]
Claim 8.16: Existence of s [proof]
Denition 8.17: 1) t is a weak -automorphism
2) x is a -task
3) t sovles x
4) x is -solvable
Observation 8.18: Basic properties
Remark 8.10: Variant of 8.20, simplifying
Main Claim 8.20: Solving one task, if possible [long proof]
Claim 8.21: Under categoricity in , etc., even M K of cardinality
(2
)
+ is
+
-saturated
Claim 8.22: variants of 8.4
Conclusion 8.23: 1) For LS(K), there is () < (2
)
+
and

or(+)
which
satises all the tasks it can [proof]
2) On bounds on ()
Denition 8.24: 1)
as in 8.23
2)
)
Remark 8.26: It was enough to prove less
9 Small pieces are enough and categoricity
Context 9.1: K categorical in ,
as in 8.23
Main Lemma = Local Lemma 9.2: For saturated M
, () < rm cf()
then every p S(M
) is
1
()-local [proof]
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

Claim 9.3: If T is cateogorical in , LS(K) () < and M
i
: i < ) is
an increasing sequence of
+
-saturated models then M
i
: i < is
+
-saturated
[CHECK:!! ()?
< cf()? quote 9.2? [proof]
Observation 9.4: 1) If M if -saturated, > rm LS(K), < , N
K
N, N K
then for some N
K
M, N
K
0
and every p S(N) realized in M is realized in
N
.
2) When increasing union is -saturated.
Theorem 9.5: If is successor,
(2
())
+ < then K is categorical in [proof]
Remark 9.6:
Assignments:
1) See 0.5(d),p.6, [Saharon?]
2) See 1.12,p.8, (ll?)
3) Concerning Denition 2.1, <
d
,
see more [Sh 600, 4], <
1
,
is called there
, cf())-brimmed over M.
4) This applies to Denition 2.8 as well.
5) PC
+
,
, 8.3??
6) 8.10 not repetition of 1 or 2. ??
7) II,1; redo for 734. !!
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

80 SAHARON SHELAH
REFERENCES.
[GrSh 238] Rami Grossberg and Saharon Shelah. A nonstructure theorem for
an innitary theory which has the unsuperstability property. Illinois
Journal of Mathematics, 30:364390, 1986. Volume dedicated to the
memory of W.W. Boone; ed. Appel, K., Higman, G., Robinson, D. and
Jockush, C.
[J] Thomas Jech. Set theory. Springer Monographs in Mathematics.
Springer-Verlag, Berlin, 2003. The third millennium edition, revised
and expanded.
[KlSh 362] Oren Kolman and Saharon Shelah. Categoricity of Theories in L
,
,
when is a measurable cardinal. Part 1. Fundamenta Mathematicae,
151:209240, 1996. math.LO/9602216.
[MaSh 285] Michael Makkai and Saharon Shelah. Categoricity of theories in L
,
with a compact cardinal. Annals of Pure and Applied Logic, 47:41
97, 1990.
[Sh 600] Saharon Shelah. Categoricity in abstract elementary classes: going up
inductively. math.LO/0011215.
[Sh:e] Saharon Shelah. Nonstructure theory, accepted. Oxford University
Press.
[Sh:F657] Saharon Shelah. On linear orders.
[Sh 3] Saharon Shelah. Finite diagrams stable in power. Annals of Mathe-
matical Logic, 2:69118, 1970.
[Sh 48] Saharon Shelah. Categoricity in
1
of sentences in L
1
,
(Q). Israel
Journal of Mathematics, 20:127148, 1975.
[Sh:a] Saharon Shelah. Classication theory and the number of nonisomor-
phic models, volume 92 of Studies in Logic and the Foundations of
Mathematics. North-Holland Publishing Co., Amsterdam-New York,
xvi+544 pp, $62.25, 1978.
[Sh 87a] Saharon Shelah. Classication theory for nonelementary classes, I. The
number of uncountable models of L
1
,
. Part A. Israel Journal of
Mathematics, 46:212240, 1983.
[Sh 87b] Saharon Shelah. Classication theory for nonelementary classes, I. The
number of uncountable models of L
1
,
. Part B. Israel Journal of
Mathematics, 46:241273, 1983.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

[Sh 88] Saharon Shelah. Classication of nonelementary classes. II. Abstract
elementary classes. In Classication theory (Chicago, IL, 1985), vol-
ume 1292 of Lecture Notes in Mathematics, pages 419497. Springer,
Berlin, 1987. Proceedings of the USAIsrael Conference on Classica-
tion Theory, Chicago, December 1985; ed. Baldwin, J.T.
[Sh 220] Saharon Shelah. Existence of many L
,
-equivalent, nonisomorphic
models of T of power . Annals of Pure and Applied Logic, 34:291
310, 1987. Proceedings of the Model Theory Conference, Trento, June
1986.
[Sh 300] Saharon Shelah. Universal classes. In Classication theory (Chicago,
IL, 1985), volume 1292 of Lecture Notes in Mathematics, pages 264
418. Springer, Berlin, 1987. Proceedings of the USAIsrael Conference
on Classication Theory, Chicago, December 1985; ed. Baldwin, J.T.
[Sh:c] Saharon Shelah. Classication theory and the number of nonisomor-
phic models, volume 92 of Studies in Logic and the Foundations of
Mathematics. North-Holland Publishing Co., Amsterdam, xxxiv+705
pp, 1990.
[Sh 394] Saharon Shelah. Categoricity for abstract classes with amalga-
mation. Annals of Pure and Applied Logic, 98:261294, 1999.
math.LO/9809197.
[Sh 620] Saharon Shelah. Special Subsets of
cf()
, Boolean Algebras and
Maharam measure Algebras. Topology and its Applications, 99:135
235, 1999. 8th Prague Topological Symposium on General Topology
and its Relations to Modern Analysis and Algebra, Part II (1996).
math.LO/9804156.
[Sh 576] Saharon Shelah. Categoricity of an abstract elementary class in two
successive cardinals. Israel Journal of Mathematics, 126:29128, 2001.
math.LO/9805146.
[Sh 472] Saharon Shelah. Categoricity of Theories in L
, when
is a measur-
able cardinal. Part II. Fundamenta Mathematicae, 170:165196, 2001.
math.LO/9604241.
[Sh:h] Saharon Shelah. Classication Theory for Abstract Elementary
Classes, volume 18 of Studies in Logic: Mathematical logic and foun-
dations. College Publications, 2009.
[ShVi 635] Saharon Shelah and Andres Villaveces. Toward Categoricity for Classes
with no Maximal Models. Annals of Pure and Applied Logic, 97:125,
1999. math.LO/9707227.
(
3
9
4
)

r
e
v
i
s
i
o
n
:
2
0
0
4
-
1
0
-
2
9

m
o
d
i
f
i
e
d
:
2
0
0
4
-
1
0
-
3
1

82 SAHARON SHELAH
[Va02] Monica M. VanDieren. Categoricity and Stability in Abstract Elemen-
tary Classes. PhD thesis, Carnegie Melon University, Pittsburgh, PA,
2002.

Saharon Shelah - Categoricity For Abstract Classes With Amalgamation

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Saharon Shelah - Categoricity For Abstract Classes With Amalgamation

Uploaded by

Copyright:

Available Formats

(

-equivalence the extension is unique). Secondly, if M

) then for some i we

-equivalent, |M| < where K is categorical in we have p

means that above we assume M

(K) = Min : is a Skolem Lowenheim number of K.

(K) |M|, then there is M

for every < (2

(i.e., in K of cardinality ) up to isomor-

) (for some but the same for = 1, 2) and M

is the identity) for = 1, 2 satisfying f

is cf()-saturated (if cf() > LS(K)).

I is nite then for some automorphism g of I we have

hence are isomorphic, so by renaming equal. So let a

and let J I, [J[ be such that M EM

. We can nd u , [u[ = such that M

(u, ). Clearly it is enough

, [S(M)[ > , there is M

(M)[ . So we have proved

is the following relation,

q i for some M K, m < we have

such that M <

) such that n < p

, then for some N we have M <

into M then h can be

for = 1, 2, h an isomorphism from M

= N and the same element realizes p by the

and cf() > then:

is (, )-brimmed if for some M

for i + 1. For each i < let N

and for each

and if we start with M

). Now for each

) is an indiscernible sequence (of character < ) (over A) if for every g,

(with domain of cardinality

) is an indiscernible set (of character < ) (over A) if: for every g, a

(with [Dom(g)[ < ) there is f AUT(C),

) is a strictly indiscernible sequence, if i

and for some , proper

which is an indiscernible sequence for quantier

and for every u [i

for arbitrarily large <

there is an isomorphism g from EM(J, ) EM(I, ) (those are (N

) is a strictly indiscernible sequence over A of character

for every > ( + LS(K))

) realizing all of them; recall is the cardinality of C.

(K)] is the set of regular such that for some

(K), e.g., 4.11(2) is not clear for

(K), = cf() then

(K) and similarly of

) is a strictly indiscernible sequence, a

, [u[ < and the isomorphism type of (C, a

proper for linear orders and a formula (not necessarily

) having the skeleton a

: < ) is strictly increasing or strictly decreasing; without loss of

(M) the set I

) : < ) satises tp( a

for = 1, 2. So for every linear order I

we can dene a linear order I

with set of elements

linearly ordered by:

, the order preserv-

: w) is indiscernible in the linear order sense (make them a