6

2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


STRONG SPLITTING IN STABLE HOMOGENEOUS MODELS
Tapani Hyttinen

and Saharon Shelah

Abstract
In this paper we study elementary submodels of a stable homogeneous structure.
We improve the independence relation dened in [Hy]. We apply this to prove
a structure theorem. We also show that dop and sdop are essentially equivalent,
where the negation of dop is the property we use in our structure theorem and sdop
implies nonstructure, see [Hy].
1. Basic denitions and spectrum of stability
The purpose of this paper is to develop theory of independence for elementary
submodels of a homogeneous structure. We get a model class of this kind if in
addition to its rst-order theory we require that the models omit some (reasonable)
set of types, see [Sh1]. If the set is empty, then we are in the classical situation from
[Sh2]. In other words, we study stability theory without the compactness theorem.
So e.g. the theory of -ranks is lost and so we do not get an independence notion
from ranks. Our independence notion is based on strong splitting. It satises the
basic properties of forking in a rather weak form. The main problem is nding free
extensions. So the arguments are often based on the denition of the independence
notion instead of the independence-calculus.
Partially supported by the Academi of Finland.
Research supported by the United States-Israel Binational Science Foundation.
Publ. 629.
1
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


Throughout this paper we assume that M is a homogeneous model of similarity
type (=language) L and that M is -stable for some < [M[ (see [Sh2] I Denition
2.2). Let (M) be the least such . By [Sh1], (M) < ((2
|L|+
)
+
). We use M
as a monster model and so we assume that the cardinality of M is large enough
for all constructions we do in this paper. In fact we assume that [M[ is strongly
inaccessible. Alternatively we could assume less about [M[ and instead of studying
all elementary submodels of M, we could study suitably small ones.
Notice Th(M) may well be unstable. Notice also that if is a stable nite
diagram, then has a monster model like M, see [Sh1].
By a model we mean an elementary submodel of M of cardinality < [M[ , we
write /, B and so on for these. So if / B are models, then / is an elementary
submodel of B Similarly by a set we mean a subset of M of cardinality < [M[ , unless
we explicitly say otherwise. We write A, B and so on for these. By a, b and so on
we mean a nite sequence of elements of M. By a A we mean a A
length(a)
.
By an automorphism we mean an automorphism of M. We write Aut(A) for
the set of all automorphisms of M such that f A = id
A
. By S

(A) we mean the

the set of all consistent complete types over A and by t(a, A) we mean the type
of a over A in M. S
m
(A) means the set t(a, A)[ a M, length(a) = m and
S(A) =
m<
S
m
(A).
We dene (M) as (T) is dened in the case of stable theories but for strong
splitting i.e. we let (M) be the least cardinal such that there are no a, b
i
and c
i
,
i < (M), such that
(i) for all i < (M), there is an innite indiscernible set I
i
over
j<i
(b
j
c
j
)
such that b
i
, c
i
I
i
,
(ii) for all i < (M), there is
i
(x, y) such that [=
i
(a, b
i
)
i
(a, c
i
).
We say that a type p over A is M-consistent if there is a M such that
p t(a, A) (i.e. there is q S(A) such that p q ).
1.1 Lemma. ([Hy]) If p S

(A) is not M-consistent, then there is nite

B A such that p B is not M-consistent.
1.2 Lemma.
(i) If (a
i
)
i<
is order-indiscernible over A then it is indiscernible over A.
(ii) Assume M is -stable and [I[ > [A[ . Then there is J I of power
> such that it is indiscernible over A.
(iii) If I is innite indiscernible over A then for all [M[ there is J I of
power such that J is indiscernible over A.
(iv) For all indiscernible I and (x, a), either X = b I[ [= (b, a) or
Y = b I[ [= (b, a) is of power < (M).
(v) There are no increasing sequence of sets A
i
, i < (M), and a such that for
all i < (M), t(a, A
i+1
) splits over A
i
. So for all A and p S(A), there is B A
of power < (M), such that p does not split over B.
(vi) For all A and p S(A), there is B A of power < (M), such that p
does not split strongly over B.
2
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


Proof. (i), (ii) and (v) as in [Hy]. (iii) follows immediately from the homogene-
ity of M. (vi) is trivial.
We prove (iv): Assume not. Let I be a counter example. Clearly we may
assume that [I[ = (M). Then By Lemma 1.1, for every J I , the type
p
J
= (b, y)[ b J (b, y)[ b I J
is M-consistent. Clearly this contradicts (M)-stability of M.
1.3 Corollary. (M) (M).
Proof. Follows immediately from Lemma 1.2 (v).
We will use Lascar strong types instead of strong types:
1.4 Denition. Let SE
n
(A) be the set of all equivalence relation E in M
n
,
such that the number of equivalence classes is < [M[ and for all f Aut(A), a E b
i f(a) E f(b). Let SE(A) =
n<
SE
n
(A).
Notice that E SE(A) need not be denable but an indiscernible set over A
is also an indiscernible set for all E SE(A).
Usually we either do not mention the arities of the equivalence relations we work
with, or we mention that the arity is f.ex. m, but we do not specify what m is.
This is harmless since usually there is no danger of confusion.
1.5 Lemma. If I is an innite indiscernible set over A, then for all E
SE(A) and a, b I , a E b.
Proof. Assume not. Let E SE(A) be a counter example. Then for all
a, b I , a ,= b, (a E b). Then Lemma 1.2 (iii) implies a contradiction with the
number of equivalence classes of E.
1.6. Lemma. If E SE(A), [A[ and M is -stable, then the number
of equivalence classes of E is .
Proof. Assume not. Then by Lemma 1.2 (ii), we can nd I such that it is
innite indiscernible over A and for all a, b I , if a ,= b then (a E b). This
contradicts Lemma 1.5.
1.7 Corollary. For all A and n < , there is E
n
min,A
SE
n
(A) such that
for all a, b and E SE
n
(A), a E
n
min,A
b implies a E b.
Proof. Clearly [SE
n
(A)[ is restricted ( 2
|S(A)|
) and so SE
n
(A) SE(A).
Trivially SE
n
(A) has the wanted property.
1.8 Denition.
(i) We say that / is F
M

-saturated if for all A / of power < and a, there

is b / such that t(b, A) = t(a, A).
(ii) We say that / is strongly F
M

-saturated if for all A / of power <

and a of length m, there is b / such that b E a for all E SE
m
(A). We write
a-saturated for strongly F
M
(M)
-saturated.
3
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


1.9 Lemma.
(i) If / is strongly F
M

-saturated then it is F
M

-saturated.
(ii) Assume [A[ , M is -stable,
<
= and there is a regular cardinal
such that . Then there is strongly F
M

-saturated / A such that

[/[ . Further more if B A is strongly F
M

-saturated, then we can choose /

so that / B.
(iii) Assume M is -stable, / is F
M

-saturated, A / is of power < and

m < . Then there are a
i
/, i < , such that for all b of length m, there is
i < such that a
i
E b, for all E SE
m
(A) i.e. / is strongly F
M

-saturated.
(iv) If / is F
M
(M)
-saturated, then it is a-saturated.
(v) Assume / is strongly F
M

-saturated and A / is of power < . Then

for all B of power < , there is f Aut(A) such that f(B) / and for all (nite
sequences) b B, f(b) E
m
min,A
b.
Proof. (i) is trivial.
(ii): For all i , choose sets A
i
of power as follows: Let A
0
= A and if
i is limit then A
i
=
j<i
A
j
. If A
i
is dened, then we let A
i+1
A
i
be such that
for all B A
i
of power < and a there is b A
i+1
such that b E
m
min,B
a. By
Lemma 1.6, we can nd A
i+1
so that [A
i+1
[ . By Lemma 1.7, A

is as wanted.
(iii): By Lemma 1.6, choose b
i
, i < , so that for all b there is i < such
that b E
m
min,A
b
i
. Since / is F
M

-saturated, we can choose a

i
/ so that there is
f Aut(A) such that for all i < , f(b
i
) = a
i
. Clearly this implies the claim.
(iv): Immediate by (iii).
(v): For all c B, choose a
c
/ so that a
c
E
m
min,A
c. Since / is F
M

-
saturated, there is f Aut(A a
c
[ c B) such that f(B) /. Clearly f is as
wanted.
1.10 Denition. We write f Saut(A) if f Aut(A) and for all a,
f(a) E
m
min,A
a.
1.11 Lemma. Assume M is -stable and [A[ < . If a E
m
min,A
b, then there
is f Saut(A) such that f(a) = b.
Proof. We dene a E b if there is f Saut(A) such that f(a) = b. Clearly it
is enough to show that E SE(A). For a contradiction, assume that this is not the
case. Since E is an equivalence relation and f(E) = E for all f Aut(A), there
are a
i
, i <
+
, such that for all i ,= j , (a
i
E a
j
). Choose B A of power such
that every E
m
min,A
-equivalence class is represented in B. Since M is -stable, there
are i < j <
+
, such that t(a
i
, B) = t(a
j
, B). Then there is f Aut(B) such that
f(a
i
) = f(a
j
). By the choice of B, f Saut(A), a contradiction.
1.12 Lemma. Assume is such that for some

, M is

-stable. If /
is F
M

-saturated and A / has power < , then t(a, /) does not split strongly
over A i for all b, c / and , b E
m
min,A
c implies [= (a, b) (a, c).
4
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


Proof. If t(a, /) splits strongly over A, then by Lemma 1.5, there are b, c /
and , such that b E
m
min,A
c and [= ((a, b) (a, c)). So we have proved the
claim from right to left. We prove the other direction: For a contradiction assume
that there are b, c / and , such that b E
m
min,A
c and [= (a, b) (a, c).
We dene an equivalence relation E on M
m
as follows: a E b if a = b or there
are I
i
, i < n < , such that they are innite indiscernible over A, a I
0
, b I
n1
and for all i < n 1, I
i
I
i+1
,= . Clearly E is an equivalence relation and for all
f Aut(A), f(E) = E. By Lemma 1.2 (ii), the number of equivalence classes of E
is < [M[ . So E SE
m
(A).
Then b E c and b ,= c. Let I
i
, i < n, be as in the denition of E. Since
/ is F
M
|A|
+
+
-saturated, we may assume that for all i < n, I
i
/. Since t(a, /)
does not split strongly over A, for all d I
0
, [= (a, d). So there is d I
1
such
that [= (a, d). Again since t(a, /) does not split strongly over A, for all d I
1
,
[= (a, d). We can carry this on and nally we get that [= (a, c), a contradiction.
1.13. Lemma. Assume A /, [A[ < (M), / is a-saturated and p S(/)
does not split strongly over A. Then for all B /, there is q S(B) such that
p q and for all C B there is r S(C), which satises q r and r does not
split strongly over A.
Proof. We dene q S

(B) as follows: (x, b) q , b B, if there is a /

such that a E
m
min,A
b and (x, a) p, where m = length(b). By Lemma 1.12, it is
enough to show that q is M-consistent. By Lemma 1.1, it is enough to show that
for all a, a

/, if a E
m
min,A
a

, then (x, a) p implies (x, a

) p. This follows
from Lemma 1.12, since by Lemma 1.9 (i), / is F
M
(M)
-saturated.
1.14 Lemma. Assume A / B, [A[ < (M), B is F
M
(M)
-saturated and
for every c B there is d / such that d E
m
min,A
c. If t(a, /) = t(b, /) and both
t(a, B) and t(b, B) do not split strongly over A, then t(a, B) = t(b, B).
Proof. For a contradiction, assume c B and [= (a, c) (b, c). Choose
d / such that d E
m
min,A
c. By Lemma 1.12, [= (a, d) (b, d), a contradiction.
1.15 Lemma. If = (M) +
<(M)
, then M is -stable.
Proof. Clearly we may assume that > (M) and so by Corollary 1.3,
(M)
+
. Let A be a set of power . We show that [S(A)[ .
Claim. There is / A such that
(i) / is F
M
(M)
-saturated,
(ii) [/[ ,
(iii) for all B / of power < (M) there is /
B
/ of power (M) satisfying:
B /
B
and for all c M there is d /
B
such that d E
m
min,A
c.
Proof. By induction on i < (M)
+
, we dene /
i
so that [/
i
[ , A /
0
,
for i < j , /
j
/
i
and
5
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


(1) if i is odd then for all B
j<i
/
j
of power < (M), there is /
B
/
i
of
power (M) satisfying: B /
B
and for all c M there is d /
B
such that
d E
m
min,A
c,
(2) if i is even then for all B
j<i
/
j
of power < (M), every p S(B) is
realized in /
i
.
By Corollary 1.3, Lemma 1.6 and the fact that [S(B)[ (M) for all B of power
< (M)
+
, it is easy to see that such /
i
, i < (M), exist. Clearly / =
i<(M)
+/
i
is as wanted. Claim.
So it is enough to show that [S(/)[ . By Lemma 1.2 (vi), for each p S(/),
choose B
p
so that p does not split strongly over B
p
and [B
p
[ < (M). Then by
Lemma 1.14, every type p S(/) is determined by p /
B
p
and the fact that it
does not split strongly over B. Since the number of possible B is
<(M)
= and
for each such B, [S(/
B
)[ (M), [S(/)[ (M) = .
1.16 Lemma. If
<(M)
> , then M is not -stable.
Proof. By the denition of (M), we may assume that (M). Let <
(M) be the least cardinal such that

> . By the denition of (M), there are

a, b
i
and c
i
, i < , such that
(i) for all i < , there is an innite indiscernible set I

i
over
j<i
(b
j
c
j
) such
that b
i
, c
i
I

i
,
(ii) for all i < , there is
i
(x, y) such that [=
i
(a, b
i
)
i
(a, c
i
).
Claim. There are I
i
, i < , such that for all i < , I
i
= d
i
k
[ k < is
indiscernible over
j<i
I
j
, b
i
, c
i
I
i
and for k < k

< , d
i
k
,= d
i
k

.
Proof. By induction on 0 < , we dene I

i
= d
,i
k
[ k < , i < , such
that
(1) for all i < , I

i
is indiscernible over
j<i
I

j
and b
i
, c
i
I

i
,
(2) for all < , there is an automorphism f such that f
j<
(b
j
c
j
) =
id

j<
(b
j
c
j
)
and for all j < , f(d
,j
k
) = d
,j
k
, k < ,
(3) for all i < and k < k

< , d
,i
k
,= d
,i
k

.
Clearly this is enough, since then I

i
, i < , are as wanted.
By (2) and homogeneity of M, limits are trivial, so we assume that = + 1
and that I

j
, j < , are dened. By Lemma 1.15, there is > such that M is -
stable. By the assumptions and Lemma 1.2 (iii), there is J = d
k
[ k <
+
such that
it is indiscernible over
j<
(b
j
c
j
) and b

, c

J . By Lemma 1.2 (ii), there is I J

of power , such that it is indiscernible over
j<
I

j
. Since J is indiscernible over

j<
(b
j
c
j
), there is an automorphism f such that f
j<
(b
j
c
j
) = id

j<
(b
j
c
j
)
and b

, c

f(d)[ d I. We let I

= f(I) and if i < , then I

i
= f(I

i
). Clearly
these are as required. Claim.
By Lemma 1.2 (iv) we may assume that for all i < , [=
i
(a, d
i
k
) i k = 0.
Then for all

and 0 < , we dene function f

so that the following

holds (f

0
= id
M
):
(a) for all i < < and

, f

I
i
= f

I
i
,
6
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


(b) if = + 1 and

, then
f

(f

(d

0
)) = f

(d

()
),
f

(f

(d

()
)) = f

(d

0
)
and for all i < , i ,= 0, (),
f

(f

(d

i
)) = f

(d

i
),
(c) if =

then f

= f

.
It is easy to see that such f

exist. For limit this follows from the homogeneity

of M and for successors this follows from the fact that f

(I

) is indiscernible over

i<
f

(I
i
).
For all

, let a

= f

(a). Then clearly for ,=

, the types of a

and a

over A = f

+1
(I

)[

, < are dierent. By the choice of ,

<
=
and so by (c), [A[ = . Since

> , M is not -stable.

So we have proved the following theorem. With slightly dierent denitions this
theorem is already proved in [Sh1].
1.17 Theorem. M is -stable i = (M) +
<(M)
.
Proof. Follows from Lemmas 1.15 and 1.16.
Let
r
(M) be the least regular (M). By Lemma 1.16, (M)
<(M)
=
(M) and so cf((M)) (M). Because cf((M)) is regular,
r
(M) (M).
2. Indiscernible sets
In this chapter we prove basic properties of indiscernible sets. We start by
improving Lemma 1.2 (iv).
2.1 Lemma. For all innite indiscernible I and a there is p S(a) such
that
[b I[ t(b, a) ,= p[ < (M).
Proof. Assume not. By Lemma 1.2 (iii), we may assume that I and a are
such that I = b
i
[ i < (M) + (M), b
i
,= b
j
for i ,= j and for some p S(a),
t(b
i
, a) = p i i (M). For all i < (M), we dene A
i
as follows:
(i) A
0
= ,
(ii) A
i+1
= A
i
b
i1
b
j
[ i j < (i + 1),
(iii) for limit i, A
i
=
j<i
A
j
.
Then it is easy to see that for all i < (M) t(a, A
i+1
) splits strongly over A
i
, a
contradiction.
2.2 Corollary. For all indiscernible I and (x, a), either X = b I[ [=
(b, a) or Y = b I[ [= (b, a) is of power < (M).
7
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


Proof. Follows immediately from Lemma 2.1.
2.3 Denition. If I is indiscernible and of power (M), we write Av(I, A)
for (x, a)[ a A, L, [b I[ [= (b, a)[ < (M).
2.4 Lemma.
(i) If I is indiscernible over A and of power (M), then Ib is indiscernible
over A i t(b, I A) = Av(I, I A).
(ii) If I and J are of power (M) and I J is indiscernible, then for all A,
Av(I, A) = Av(J, A).
(iii) If I is indiscernible and of power (M), then for all A, Av(I, A) is
M-consistent.
Proof. (i) and (ii) are trivial. We prove (iii): By (ii) and Lemma 1.2 (iii), we
may assume that [I[ > [L A[ +
r
(M). Then the claim follows by the pigeon hole
principle from (i).
2.5 Denition. Assume I and J are indiscernible sets of power (M).
(i) We say that I is based on A if for all B A I , Av(I, B) does not split
strongly over A.
(ii) We say that I and J are equivalent if for all B, Av(I, B) = Av(J, B)
(iii) We say that I is stationary over A if I is based on A and for all f
Aut(A), f(I) and I are equivalent.
2.6 Lemma. Assume I is an indiscernible set of power (M), [A[ < and
M is -stable. Then the following are equivalent:
(i) I is based on A,
(ii) the number of non-equivalent indiscernible sets in f(I)[ f Aut(A) is
,
(iii) the number of non-equivalent indiscernible sets in f(I)[ f Aut(A) is
< [M[ .
Proof. (i)(ii): Assume not. Let f
i
(I), i <
+
, be a counter example. For
all i < (M), choose /
i
so that
(a) A /
0
and every type p S(A) is realized in /
0
,
(b) if i < j , then /
i
/
j
and for limit i, /
i
=
j<i
/
j
,
(c) every type p S(/
i
) is realized in /
i+1
,
(d) [/
i
[ .
Let / =
i<(M)
/
i
. Since M is -stable there are i ,= j such that Av(f
i
(I), /) =
Av(f
j
(I), /). Let a be such that Av(f
i
(I), / a) ,= Av(f
j
(I), / a). By
Lemma 1.2 (v), choose i < (M) so that t(a, /
i+
) does not split over /
i
. Without
loss of generality, we may assume that i = 0. For all i < , choose a
i
/
i+1
so that
t(a
i
,
ji
/
j
) = t(a,
ji
/
j
). By an easy induction, we see that a a
i
[ i < is
order-indiscernible over / and so also over A. By Lemma 1.2 (i), aa
i
[ i < is
indiscernible over A. But then clearly either Av(f
i
(I), /a) or Av(f
j
(I), /a)
splits strongly over A, a contradiction.
8
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


(ii)(iii): Trivial.
(iii)(i): Assume not. Then by Lemma 1.2 (iii), we can nd J = a
i
[ i < [M[
and (x, y) such that J is indiscernible over A, for i ,= j , a
i
,= a
j
, and (x, a
i
)
Av(I, J) i i = 0. But then for all i < [M[ , we can nd f
i
Aut(A) such that for
all j < i, (x, a
j
) , Av(f
i
(I), J) but (x, a
i
) Av(f
i
(I), J). Clearly these f
i
(I)
are not equivalent, a contradiction.
3. Independence
In this chapter we dene an independence relation and prove the basic properties
of it. This independence notion is an improved version of the one dened in [Hy]. It
satises weak versions of the basic properties of forking. E.g. a
A
A holds assuming
A is a-saturated.
3.1 Denition.
(i) We write a
A
B if there is C A of power < (M) such that for all
D AB there is b which satises: t(b, AB) = t(a, AB) and t(b, D) does not
split strongly over C. We write C
A
B if for all a C, a
A
B.
(ii) We say that t(a, A) is bounded if [b[ t(b, A) = t(a, A)[ < [M[ . If t(a, A)
is not bounded, we say that it is unbounded.
3.2 Lemma.
(i) If A A

B and a
A
B then a
A
B

.
(ii) If A B and a
A
B then for all C B there is b such that t(b, B) =
t(a, B) and b
A
C.
(iii) Assume that / is a-saturated. If A / is such that t(a, /) does not split
strongly over A then for all B such that A B /, a
B
/. Especially a
A
/.
(iv) Assume a and A are such that t(a, A) is bounded. Then for all B A,
t(a, B) does not split strongly over A.
(v) Assume A B and t(a, A) is unbounded. If t(a, B) is bounded, then
a ,
A
B.
(vi) Assume / is a-saturated and a , /. Then t(a, /) is unbounded.
(vii) Let be a cardinal. Assume a and A are such that t(a, A) is unbounded
and a
A
A. If a
i
, i < , are such that for all i < , t(a
i
, A) = t(a, A) and
a
i

A

j<i
a
j
, then [a
i
[ i < [ = .
(viii) Assume A B, a
A
A and t(a, A) is unbounded. Then there is b such
that b
A
B and b E
m
min,A
a.
(ix) If a
A
b c and b E
m
min,A
c, then t(b, A a) = t(c, A a).
Proof. (i) is immediate by the denition of .
(ii): Choose a-saturated T C. Since a
A
B, there are b and A

A such
that t(b, B) = t(a, B), [A

[ < (M) and t(b, T) does not split strongly over A

. By
Lemma 1.13, b is as wanted.
(iii): By Lemmas 1.2 (vi) and 1.13, a
A
/ and so by (i), a
B
/.
(iv): Assume not. Then there are distinct a
i
, i < [M[ , and , such that
a
i
[ i < [M[ is indiscernible over A and [= (a, a
i
) i i = 0. For all (M) i <
9
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


[M[ , nd an automorphism f
i
Aut(A) such that f
i
(a
0
) = a
i
, f(a
i
) = a
0
and for
all 0 < j < i, f
i
(a
j
) = a
j
. By Corollary 2.2, it is easy to see that f
i
(a)[ (M)
i < [M[ contains [M[ distinct elements, a contradiction.
(v): Assume not. Then by (ii) we can nd C B and b such that t(b, B) =
t(a, B), b
A
C and b C. By Lemma 1.2 (ii), there is an innite indiscernible set
I over A such that b I . Clearly we cannot nd c such that t(c, C) = t(b, C) and
t(c, C I) does not split strongly over some A

A, a contradiction.
(vi): Follows immediately from (iii) and (v).
(vii): Immediate by (v).
(viii): Let > [A[ be such that M is -stable. Choose a
i
, i <
+
so that
t(a
i
, A) = t(a, A) and a
i

A

j<i
a
j
. By (vii) and Lemma 1.2 (ii), we may assume
that a
i
[ i < is innite indiscernible over A. Clearly we may also assume that
a = a
0
. Let d = a
1
. Then t(d, A) = t(a, A), d
A
a and by Lemma 1.5, d E
m
min,A
a.
Then we can choose b so that t(b, Aa) = t(d, Aa) and b
A
a B. Clearly then
b is as wanted.
(ix) Follows immediately from Lemma 1.12. Notice that if b E
m
min,A
c, then for
all d A, b d E
m+k
min,A
c d.
3.3 Denition.
(i) We say that M-consistent p S(A) is stationary if for all a, b and B A
the following holds: if t(a, A) = t(b, A) = p, a
A
B and b
A
B then t(a, B) =
t(b, B).
(ii) We say that I is A-independent if for all a I , a
A
I a.
3.4 Lemma. If / is a-saturated, then every M-consistent p S(/) is
stationary.
Proof. Assume not. Choose B /, a and b so that t(a, /) = t(b, /), a
A
B,
b
A
B and t(a, B) ,= t(b, B). By Lemma 3.2 (ii) we may assume that B is F
M
(M)
-
saturated. Choose c B and so that [= (a, c) (b, c). Let A / be such
that [A[ < (M) and both t(a, B) and t(b, B) do not split strongly over A. Choose
d / so that d E
m
min,A
c. By Lemma 1.12, a contradiction follows.
3.5 Corollary.
(i) Assume / is a-saturated. If a ,
A
B, then there is b B such that
a ,
A
b.
(ii) If / is a-saturated and a
i
, i < , are such that a
0
, /, for all i, j ,
t(a
i
, /) = t(a
j
, /) and a
i

A

j<i
a
j
, then a
i
[ i < is indiscernible over / and
/-independent and if i ,= j , then a
i
,= a
j
.
(iii) Assume / is a-saturated. Then for all B / and C there is D such that
t(D, /) = t(C, /) and D
A
B.
(iv) If A B C, B is a-saturated, a
A
B and a
B
C, then a
A
C.
(v) Assume / is a-saturated, t(a, /) does not split strongly over A / and
[A[ < (M). Then a ,
A
B i there is nite b / B such that a ,
A
b.
10
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


Proof. (i) follows immediately from Lemma 3.4 (if a ,
A
B, then t(a, / B)
is not the unique free extension of t(a, /), which can be detected from a nite
sequence).
(ii): By Lemma 3.4, it is easy to see that a
i
[ i < is order-indiscernible over
/. By Lemma 1.2 (i), a
i
[ i < is indiscernible over /. Clearly this implies that
a
i
[ i < is /-independent. The last claim follows from Lemma 3.2 (v).
(iii): Clearly it is enough to prove the following: If D
A
B, then for all c there
is d such that t(d, /D) = t(c, /D) and dD
A
B. This follows from Lemmas
1.1, 3.2 (ii) and 3.4.
(iv): Choose b so that t(b, B) = t(a, B) and b
A
C. Then b
B
C and so by
Lemma 3.4, we get t(b, C) = t(a, C). Clearly this implies the claim.
(v): If a
A
B then by (iv), a
A
/ B from which it follows that there
are no nite b / B such that a ,
A
b. On the other hand if a ,
A
B, then
t(a, /B) is not the unique free extension of t(a, /) dened in the proof of Lemma
1.13. This means that there are c / and d / B such that c E
m
min,A
d and
t(c, A a) ,= t(d, A a). Clearly a ,
A
c d.
3.6 Lemma. If / is a-saturated and a
A
b, then b
A
a.
Proof. Assume not. Let > [/[ be such that M is -stable. For all i <
+
,
choose a
i
and b
i
so that t(a
i
, /) = t(a, /), a
i

A

j<i
(a
j
b
j
), t(b
i
, /) = t(b, /)
and b
i

A
a
i

j<i
(a
j
b
j
). Then by Lemma 3.4, b
i
,
A
a
j
i j > i. Clearly this
contradicts Lemma 1.2 (ii).
3.7 Corollary. For all a, b and A, b
A
A and a
A
b implies b
A
a.
Proof. Assume not. Choose a-saturated / A and b

so that t(b

, A) =
t(b, A) and b

A
/. We may assume that b

= b. Then choose a

so that t(a

.A
b) = t(a, A b) and a

A
/ b. By Lemma 3.6, b
A
a

. By Corollary 3.5 (iii),

b
A
a

and so b
A
a.
3.8 Lemma.
(i) If b
A
D and c
Ab
D, then b c
A
D.
(ii) If / is a-saturated, B
A
D and C
AB
D, then B C
A
D.
(iii) Assume / is a-saturated and B /. If a
A
B, a
B
C and there is
D B (f.ex. D = B) such that C
D
B, then a
A
B C.
(iv) Assume / is a-saturated. If a
A
b and a b
A
B, then a
A
B b.
(v) Assume a
A
A, for all i < , t(a
i
, A) = t(a, A) and a
i

A

j<i
a
j
. Then
for all n < , a
i
[ i < n is A-independent.
Proof. (i): Choose B A of power < (M) such that
(a) for all C AD there is b

which satises: t(b

, AD) = t(b, AD) and

t(b

, C) does not split strongly over B

and
(b) for all C ADb there is c

which satises: t(c

, ADb) = t(c, ADb)

and t(c

, C) does not split strongly over B b.

11
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


Let C AD be arbitrary. Choose b

as in (a) above. By (b) above we can nd c

such that t(c

, A D) = t(c b, A D) and t(c

, C b

) does not split strongly

over B b

.
For a contradiction, assume t(b

, C) splits strongly over B. Let I = a

i
[ i <
C and be such that I is indiscernible over B and
(c) [= (c

, b

, a
0
) (c

, b

, a
1
).
Claim. I is indiscernible over B b

.
Proof. If not, then (change the enumeration if necessary) there is over B
such that [= (b

, a
0
, ..., a
n1
) (b

, a
n
, ..., a
2n1
). Since
(a
mn
, ..., a
(m+1)n1
)[ m <
is indiscernible over B, we have a contradiction with the choice of b

. Claim.
By Claim and (c), t(c

, C b

) splits strongly over B b

. This contradicts the

choice of c

.
(ii): Clearly we may assume that C is nite. Let b B be arbitrary. We show
that C b
A
D. Choose A / and A

B such that
(a) b A

, [A A

[ < (M),
(b) for all D

/ B D there is C

which satises: t(C

, / B D) =
t(C, A B D) and t(C

, D

) does not split strongly over A A

(c) for all D

/ D and a A

, there is a

which satises: t(a

, / D) =
t(a, A D) and t(a

, D

) does not split strongly over A.

Then we can proceed as in (i). (We assume that / is a-saturated in order to be
able to use Corollary 3.5 (iii).)
(iii): By Lemma 3.6, B
A
a. By Corollary 3.7, C
B
a. By (ii), these imply
B C
A
a, from which we get the claim by Lemma 3.6.
(iv): Choose a

so that t(a

, / b) = t(a, / b) and a

A
B b. By (i) and
Lemma 3.4, t(a

b, / B) = t(a b, / B).
(v): By (i) it is easy to see that
(*) for all n < ,
i<n
a
i

A
A.
We prove the claim by induction on n. For n = 1 the claim follows immediately
from the assumptions. Let i < n. We show that a
i

A
a
j
[ j < n, j ,= i. If
i = n1, then this is assumption. So assume that i < n1. By the choice of a
n1
,
a
n1

A

{a
j
| j<n1, j=i}
a
i
.
By the induction assumption
a
i

A
a
j
[ j < n 1, j ,= i
and by (*) and Corollary 3.7
a
j
[ j < n 1, j ,= i
A
a
i
.
By (i),
a
n1

a
j
[ j < n 1, j ,= i
A
a
i
.
By Corollary 3.7, the claim follows.
12
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


3.9 Lemma. Assume B A and t(a, A) is unbounded. Then a
A
B i
there is an indiscernible set I over A such that [I[ (M), I is based on some
A

A of power < (M) and Av(I, B) = t(a, B).

Proof. From right to left the claim is trivial. So we prove the other direction.
Without loss of generality, we may assume that B is a-saturated. Let A

A be
such that [A

[ < (M) and for all C B there is b such that t(b, B) = t(a, B) and
t(b, C) does not split strongly over A

. Let > [B[ be a regular cardinal such that

M is -stable. For all i <
+
we dene B
i
and a
i
so that
(i) B
i
, i <
+
, is an increasing sequence of -saturated models of power and
B B
0
,
(ii) for all i <
+
, t(a
i
, B) = t(a, B), a
i
B
i+1
B
i
and t(a
i
, B
i
) does not
split strongly over A

(so a
i

A
B
i
).
By Lemma 3.2 (v) and Corollary 3.5 (ii), a
i
[ i <
+
is indiscernible over B and
a
j
,= a
j
for all i < j <
+
. We prove that I = a
i
[ i < (M) is as wanted.
Clearly it is enough to show that I is based on A

. For a contradiction, assume

that C B is such that Av(I, C) splits strongly over A

. Clearly we may assume

that C B
(M)+1
. By Lemma 1.2 (ii) there is J
+
((M) + 1), such that
[J[ =
+
and a
i
[ i J is indiscernible over C. Then t(a
i
, C) = Av(I, C) for
all i J . By (ii) above, for all i J , t(a
i
, C) does not split strongly over A

, a
contradiction.
3.10 Lemma. Assume a E
m
min,A
b, a
A
c and b
A
c. If c
A
A or t(a, A)
is bounded or t(c, A) is bounded, then t(a, A c) = t(b, A c).
Proof. We divide the proof to three cases:
Case 1. t(c, A) is bounded: Let B be the set of all e such that t(e, A) is
bounded. Then [B[ < [M[ and so [S(AB)[ < [M[ . We dene E so that x E y if
t(x, AB) = t(y, AB). Since for all f Aut(A), f(AB) = AB, E SE(A).
Clearly this implies the claim.
Case 2. t(a, A) is bounded: Dene E so that x E y if x = y or t(x, A) ,= t(a, A)
and t(y, A) ,= t(a, A). Clearly E SE
m
(A), and so a = b from which the claim
follows.
Case 3. t(a, A) is unbounded and c
A
A: Assume the claim is not true. Let
> [A[ be such that M is -stable. Choose a
i
, i <
+
so that t(a
i
, A c) =
t(a, Ac) and a
i

A
c

j<i
a
j
. By Lemmas 3.2 (vii) and 1.2 (ii), we may assume
that a
i
[ i < is innite indiscernible over A. Clearly we may also assume that
a = a
0
. Let d = a
1
. Then t(d, A c) = t(a, A c), d
A
a c and by Lemma
1.5, d E
m
min,A
a. Then we can choose this d so that in addition, d
A
a c b. By
Lemma 3.8 (i), b d
A
c. By Corollary 3.7, c
A
b d. Since d E
m
min,A
b, this
contradicts Lemma 3.2 (ix).
Notice that in the case(s) 1 (and 2) above the assumptions a
A
c and b
A
c
are not used.
3.11 Corollary. Assume a
i
, i < , are such that for all i, j < , a
i
E
m
min,A
a
j
13
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


and for all i < , a
i

A

j<i
a
j
. Then for all i ,= j , a
i
,= a
j
and a
i
[ i < is
indiscernible over A.
Proof. By Lemma 3.2 (vii), for all i ,= j , a
i
,= a
j
. We show that for all
i
0
< i
1
< ... < i
n
< , t(a
0
... a
n
, A) = t(a
i
0
... a
i
n
, A). By Lemma 1.2 (i),
this is enough.
By Lemma 3.8 (v), a
i
[ i i
n
is A-independent and by Lemma 3.8 (i), it is
easy to see that a
i
[ i i
n

A
A. So by Lemma 3.10, t(a
0
, A

0<kn
a
i
k
) =
t(a
i
0
, A

0<kn
a
i
k
). So it is enough to show that t(a
0
... a
n
, A) = t(a
0
a
i
1

... a
i
n
, A). As above we can see that t(a
1
, A a
0

1<kn
a
i
k
) = t(a
i
1
, A a
0

1<kn
a
i
k
). So it is enough to show that t(a
0
... a
n
, A) = t(a
0
a
1
a
i
2
...
a
i
n
, A). We can carry this on and get the claim.
3.12 Theorem. Assume a
A
c, b
A
c and a E
m
min,A
b. Then t(a, A c) =
t(b, A c).
Proof. Assume not. As in the proof of Lemma 3.10 (Case 3.), we can nd a

and b

such that t(a

, A c) = t(a, A c), t(b

, A c) = t(b, A c), a

A
c a,
b

A
c b, a

E
m
min,A
a and b

E
m
min,A
b. For all i < (M), choose a
i
so that
a
i

A
c a b

j<i
a
j
, if i is odd, then t(a
i
, Ac a) = t(a

, Ac a) and if i is
even, then t(a
i
, Ac b) = t(b

, Ac b). By Corollary 3.11, for all i ,= j , a

i
,= a
j
and a
i
[ i < (M) is indiscernible over A. Clearly this contradicts Lemma 2.1.
3.13 Lemma. Assume M is -stable and [A[ . Then there is a-saturated
/ A of power .
Proof. Immediate by Lemma 1.9 (ii) and the fact that
r
(M) (M) is
regular.
3.14 Theorem. Assume M is -stable and [A[ . Then there is F
M

-
saturated / A of power .
Proof. By Lemma 3.13, there is an increasing continuous sequence A
i
, i ,
of models of power such that
(i) A A
0
and for all i , A
i+1
is a-saturated,
(ii) for all i < and a, there is b A
i+1
such that t(b, A
i
) = t(a, A
i
).
We show that / = A

is as wanted. For this let B / of power < and b be

arbitrary. We show that t(b, B) is realized in /.
By Theorem 1.17, cf()
r
(M) and so / is a-saturated and there is

<
such that b
A

/. By the pigeon hole principle there is < such that

and (A
(+1)
A

) B = .
Claim. There is < such that B
A
+
A
++1
.
Proof. Assume not. Then by the pigeon hole principle, we can nd c B such
that
[ < [ c ,
A
+
A
++1
[ cf().
But this is impossible by Lemma 3.2 (iii), because cf()
r
(M) and /

is
a-saturated for all such that cf()
r
(M). Claim.
14
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


Choose c A
++1
so that t(c, A
+
) = t(b, A
+
). By Claim, B
A
+
c and so c
A
+
B. Since b
A
+
B, Lemma 3.4 implies, t(c, A
+
B) =
t(b, A
+
B).
We nish this chapter by proving that over F
M
(M)
-saturated models our inde-
pendence notion is equivalent with the notion used in [Hy].
3.15 Lemma. Assume / is F
M
(M)
-saturated model and B /. Then the
following are equivalent:
(i) a
A
B.
(ii) For all b B there is A / of power < (M) such that t(a, / b) does
not split over A.
Proof. Let p S(/) be arbitrary M-consistent type. Let a be such that
t(a, /) = p and a
A
B. Let a

be such that t(a

, /) = p and for all b B there is

A / of power < (M) such that t(a

, /b) does not split over A. We show that

then t(a, B) = t(a

, B). This implies the claim, since for all M-consistent p S(/)
such a and a

exist: The existence of a follows from Lemma 3.2 (ii) and (iii) and
the existence of a

can be seen as in [Hy].

For a contradiction, assume that there is b B such that t(a, /b) ,= t(a

, /
b). By the choice of a and a

and Lemma 1.2 (vi), there is A / of power < (M)

such that t(a, / b) does not split strongly over A, t(a

, / b) and t(b, /) do
not split over A and t(a, A b) ,= t(a

, A b). For all i < , choose b

i
/ so
that t(b
i
, A

j<i
b
j
) = t(b, A

j<i
b
j
). Since t(b, /) does not split over A, by
Lemma 1.2 (i), it is easy to see that b
i
[ i < b is innite indiscernible over
A. Since t(a, /) = t(a

, /), either t(a, /b) or t(a

, /b) splits strongly over A,

a contradiction.
4. Orthogonality
In this chapter we study orthogonality. Since we do not have full transitivity of
, we need stationary pairs:
4.1 Denition. Assume A B and p S(B). We say that (p, A) is
stationary pair if for all a, t(a, B) = p implies a
A
B and for all C B, a
and b, the following holds: if a
A
C, b
A
C and t(a, B) = t(b, B) = p, then
t(a, C) = t(b, C).
4.2 Lemma.
(i) Assume A B C, a
A
C and (t(a, B), A) is a stationary pair. Then
(t(a, C), A) is a stationary pair.
(ii) Assume A B C D, a
A
C, a
B
D and (t(a, C), B) is a stationary
pair. Then a
A
D.
Proof. (i) is trivial, so we prove (ii): Choose a

so that t(a

, C) = t(a, C) and
a

A
D. Then a

B
D and so t(a

, D) = t(a, D) from which the claim follows.

15
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


4.3 Lemma. Assume / is a-saturated, t(a, /) does not split strongly over
A / and [A[ < (M). Then there is B / such that A B, [B A[ < ,
B
A
A and (t(a, B), A) is a stationary pair.
Proof. By Lemma 1.13, a
A
/. Choose b
i
, i , so that for all i ,
t(b
i
, /) = t(a, /) and b
i

A
/

j<i
b
j
. Then b
i
[ i is indiscernible over A
and by Lemma 3.8 (ii),
() b
i
[ i <
A
/.
Especially,
() b
i
[ i <
A
A.
Without loss of generality, we may assume that b

= a. Choose a

/ so that
a

E
m
min,A
a. Let B = A a

and I = b
i
[ i < . Then B
A
A.
Claim. Assume J I is indiscernible over A, t(b, B) = t(a, B) and b
A
B J a. Then J b is indiscernible over A.
Proof. By Lemmas 1.12 and 1.5 it is enough to show that t(b, AI) = t(a, A
I). By (), I
A
a

. By the choice of a

, a

A
A and so by Corollary 3.7,
a

A
I . By the choice of b and Lemma 3.2 (i), b
Aa
I . By Lemma 3.8 (i),
b a

A
I . By () and Corollary 3.7, I
A
a

b. So by Lemma 3.2 (ix),

t(a

, A I) = t(b, A I). Similarly we can see that I

A
a

a and so by Lemma
3.2 (ix), t(a

, A I) = t(a, A I). Claim.

We show that (t(a, B), A) is a stationary pair. Assume not. Since / is F
M
(M)
-
saturated, we can nd b such that b
A
/, t(b, B) = t(a, B) and t(b, /) ,= t(a, /).
Choose c
i
, i < (M), so that for all i < (M), t(c
i
, /) = t(b, /) if i is odd,
t(c
i
, /) = t(a, /) if i is even and for all i < (M), c
i

A
/I

j<i
c
j
. By Claim
c
i
[ i < (M) is indiscernible. This contradicts Corollary 2.2.
4.4 Denition.
(i) We say that p S(A) is orthogonal to q S(C) if for all a-saturated
/ A C the following holds: if t(b, C) = q , b
C
/, t(a, A) = p and a
A
/,
then a
A
b. We say that p S(A) is orthogonal to C if it is orthogonal to every
q S(C).
(ii) We say that a stationary pair (p, A) is orthogonal to q S(C) if for all a-
saturated / Cdom(p) the following holds: if t(b, C) = q , b
C
/, t(a, dom(p)) =
p and a
A
/, then a
A
b. We say that a stationary pair (p, A) is orthogonal to
C if it is orthogonal to every q S(C).
4.5 Lemma. Assume / is a-saturated, A B /, a
A
/ and (t(a, B), A)
is a stationary pair. Then t(a, /) is orthogonal to C i (t(a, B), A) is orthogonal
to C.
Proof. Immediate.
4.6 Lemma. Assume A /, / is a-saturated and p S(/). Then the
following are equivalent.
(i) p is orthogonal to A.
(ii) For all a and b, if t(a, /) = p and b
A
/, then a
A
b.
16
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


Proof. Clearly (i) implies (ii) and so we prove the other direction. Assume (ii)
and for a contradiction assume that there is a-saturated ( / and a and b such
that t(a, /) = p, a
A
(, b
A
( and a ,
C
b.
Choose B
0
B
1
/ so that
(1) [B
1
[ < (M),
(2) a
B
0
/ and b
B
0
A
/,
(3) (t(a, B
1
), B
0
) is a stationary pair.
By Corollary 3.5 (v), choose nite d ( such that a ,
B
1
db. Choose B
2
B
1
d
of power < (M) such that B
2
( and t(a b, () does not split strongly over B
2
.
Since t(a, () and t(b, () do not split strongly over B
2
we can nd by Lemmas 4.3
and 4.2 (i) B
3
B
2
of power < (M) such that B
3
( and both (t(a, B
3
), B
2
)
and (t(b, B
3
), B
2
) are stationary pairs. Then
(*) a
B
0
B
3
and b
B
0
A
B
3
.
Choose f Aut(B
1
) so that f(B
3
) / and for all c B
3
, f(c) E
m
min,B
1
c.
Then t(f(a), f(B
3
)) = t(a, f(B
3
)) and so we may assume that f(a) = a. Now
a f(b)
f(B
2
)
f(B
3
), and so we can nd a

and b

so that t(a

, f(B
3
)) =
t(a f(b), f(B
3
)) and a

f(B
2
)
/. Then by (*) and Lemma 4.2 (ii), a

B
0
/
and so t(a

, /) = t(a, /) and we may assume that a

= a. Also by Lemma 4.2 (ii)

and (*), b

B
0
A
/ and so b

A
/. Because a ,
B
1
f(c) b

, by Corollary 3.5 (v),

a ,
A
b

. Clearly this contradicts (ii).

4.7 Lemma. Let
r
(M) be a cardinal. Assume D C, p S(C),
(p, D) is a stationary pair and orthogonal to /, [C[ < , / B are strongly
F
M

-saturated and C
A
B. Then (p, D) is orthogonal to B.
Proof. For a contradiction, assume that q S(B) is not orthogonal to (p, D).
Choose B B of power < (M) so that q does not split strongly over B. Choose
A / so that
(i) [A[ < ,
(ii) for all c C, t(c, / B) does not split strongly over A.
By Lemma 1.9 (v), we can nd B

/ and f Aut(A) so that f(B) = B

and for
all b B, b E
m
min,A
f(b). By Lemma 1.12, t(B

, C) = t(B, C). Let q

= f(q) B

.
Then it is easy to see that q

and (p, C) are not orthogonal, a contradiction.

4.8 Corollary. Assume / B ( are strongly F
M

r
(M)
-saturated, B
A
(
and p S(() is orthogonal to /. Then p is orthogonal to B.
Proof. Follows immediately from Lemma 4.3, 4.5 and 4.7.
5. Structure of s-saturated models
We say that M is superstable if (M) = .
5.1 Lemma. The following are equivalent.
(i) (M) = .
(ii) There are no increasing sequence /
i
, i < , of a-saturated models and a
such that for all i < , a ,
A
i
/
i+1
.
17
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


(iii) There are no increasing sequence /
i
, i < , of F
M
(M)
-saturated models
and a such that for all i < , a ,
A
i
/
i+1
.
Proof. Clearly (i) implies (ii) and (ii) implies (iii). So we assume that (i) does
not hold and prove that (iii) does not hold either. For this, choose an increasing
sequence of regular cardinals
i
, i < , such that for all i < , M is
i
-stable. Let
= sup
i<

i
. By Theorem 1.17, M is not -stable. Let A be such that [A[
and [S(A)[ > . Then choose an increasing sequence /
i
, i < , of F
M
(M)
-saturated
models of power
i
such that A
i<
/
i
. Then [S(
i<
/
i
)[ > . By Corollary 3.5
(i), it is enough to show that there is a such that for all i < , a ,
A
i

i<
/
i
. For a
contradiction, assume not. Then for all a there is i
a
< , such that a
A
i
a

i<
/
i
.
Then by Lemma 3.4, for all a, t(a,
i<
/
i
) is determined by t(a, /
i
a
). Since for
all i < , [S(/
i
)[ , this implies that [S(
i<
/
i
)[ , a contradiction.
5.2 Denition. We say that t(a, A) is F
M

-isolated if there is B A of
power < , such that for all b, t(b, B) = t(a, B) implies t(b, A) = t(a, A). We
dene F
M

-construction, F
M

-primary etc, as in [Sh2]. Instead of F

M
(M)
-saturated,
F
M
(M)
-isolated etc, we write s-saturated, s-isolated etc.
In slightly dierent context, the following theorem is proved in [Sh1].
5.3 Theorem. ([Sh1]) Assume (M).
(i) For all A there is an F
M

-primary model over A.

(ii) If / is F
M

-primary over A then it is F

M

-prime over A.
(iii) If / is F
M

-primary over A and (M) is regular, then / is F

M

-atomic
over A.
(iv) If (M) is regular, then F
M

-primary models over any set A are unique

up to isomorphism over A.
As usual we write A
C
B if for all a, a
C
A implies a
C
B.
5.4 Lemma.
(i) Assume / is s-saturated and B is s-primary over / B. Then B
A
B.
(ii) Assume / B C, / is s-saturated and B
A
C. If (B, b
i
[ i <
, (B
i
[ i < )) is an s-construction over B, then (B C, b
i
[ i < , (B
i
[ i < ))
is an s-construction over B C.
(iii) Assume (M), / is F
M

-saturated and B is F
M

-primary over /B.

Then B
A
B.
Proof. (i): Assume not. Then we can nd s-saturated /, B, b and a so that
t(b, / B) is s-isolated, a
A
B and a ,
A
b (if (/ B, b
i
[ i < , (B
i
[ i < ))
is an s-construction of B, then let b = b
i
, where i is the least ordinal such that
a ,
A
B

ji
b
j
and rename B

j<i
b
j
as B; i exists by Corollary 3.5 (v)).
Without loss of generality we may assume that [B[ < (M). Choose A / so that
(i) t(b, A B) s-isolates t(b, / B),
(ii) for all c B, t(c, /a) does not split strongly over some A

A of power
< (M),
18
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


(iii) t(b, /) does not split strongly over some A

A of power < (M),

(iv) [A[ < (M).
This is possible since
r
(M) (M): Let = [B[ + 1 < (M). Clearly we can
choose A so that it of the form A

where A

is of power < (M) and A

is a
union of many sets of power <
r
(M) (M). If (M) is regular, then clearly
[A[ < (M). Otherwise
r
(M) < (M) in which case [A[ [A

[ +max(,
r
(M)) <
(M).
By Lemma 1.9 (iii), the proof of Lemma 1.13 and (iii) above, there are c, c

, a

/ such that c a E
m
min,A
c

and t(b c a, A) ,= t(b c

, A). By (ii),
t(B c a, A) = t(B c

, A). So there is f Aut(A B) such that f(c) = c

and f(a) = a

. Then f(b) contradicts (i) above.

(ii): As (i) above.
(iii): By (i) we may assume that > (M). For a contradiction, assume that
the claim does not hold. As in (i), we can nd s-saturated /, B, b and a so that
t(b, /B) is F
M

-isolated, a
A
B, a ,
A
b and [B[ < . Let A / be such that
t(a, AB) F
M

-isolates t(b, /B). Choose s-saturated ( / so that [([ = (M)

and a
C
/ B. For i < , choose a
i
/ such that (a
i
)
i<
is (-independent and
for all i < , t(a
i
, () = t(a, (). As in (i), it is enough to show that there is i <
such that a
i

C
AB. For this we choose maximal sequence of models /
j
and sets
I
j
, j j

, such that
(a) /
0
= ( and I
0
= ,
(b) I
j+1
I
j
is nite, /
j+1
is s-primary over /
j
(I
j+1
I
j
) and for some
c A B, c ,
A
j
I
j+1
I
j
,
(c) if j is limit, then I
j
=
k<j
I
k
and /
j
is s-primary over
k<j
/
k
.
Since
r
(M) [A B[ < , I
j
,= . Let i I
j
. By (i) and (ii), it is easy to
see that for all j j

, /
j
is s-primary over / I
j
. Then by (i), a
i

C
/
j
and
because the sequence was maximal, A B
A
j

a
i
. So a
i

C
A B as wanted.
5.5 Corollary.
(i) Assume A / and / is s-saturated. If p S(/) is orthogonal to A, then
for all C /, a and b the following holds: if a
A
C, t(a, /) = p and b
A
C,
then a
A
C b.
(ii) Assume M is superstable and is a limit ordinal. Let /
i
, i < , be an
increasing sequence of s-saturated models and / be s-primary over
i<
/
i
. If
a , / then there is i < such that t(a, /) is not orthogonal to /
i
.
(iii) Assume / is s-saturated and p S(/) is orthogonal to A /. If a
i
,
i < , are such that for all i < , t(a
i
, /) = p and a
i

A

j<i
a
j
, then for all
n < , t(
i<n
a
i
, /) is orthogonal to A.
Proof. (i): Assume not. Let ( be s-primary over /C. Then by Lemma 5.4
(i) and Corollary 3.5 (iv), a
A
(, b
A
( and a ,
C
b, a contradiction.
(ii): Clearly we may assume that if i < j then /
i
,= /
j
. Since (M) = ,
there is i < such that a
A
i

j<
/
j
. By (i), a
A
i
/. By Lemma 3.2 (v), this is
more that required.
19
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


(iii): Assume not. Then by Lemma 4.6, there is b such that b
A
B and

i<n
a
i
,
A
b. Let m n be the least such that
i<m
a
i
,
A
b. By Lemma 3.8 (i),
a
m1
,
A

i<m1
a
i
b. Clearly this contradicts (i).
Let P be a tree without branches of length > . Then by t

we mean the
immediate predecessor of t if t P is not the root. For all t P , by t
1
>
we mean
the set of immediate successors of t.
5.6 Denition. ([Sh2]) We say that (P, f, g) = ((P, ), f, g) is an s-free tree
of s-saturated / if the following holds:
(i) (P, ) is a tree without branches of length > , f : (P r) / and
g : P P(/), where r P is the root of P and P(/) is the power set of / - in
order to simplify the notation we write a
t
for f(t) and /
t
for g(t),
(ii) /
r
is s-primary model (over ),
(iii) if t is not the root and u

= t then t(a
u
, /
t
) is orthogonal to /
t
,
(iv) if t = u

then /
u
is s-primary over /
t
a
u
,
(v) Assume T, V P and u P are such that
(a) for all t T , t is comparable with u,
(b) T is downwards closed.
(c) if v V then for all t such that v _ t ~ u, t , T .
Then

tT
/
t

A
u

vV
/
v
.
5.7 Denition. We say that (P, f, g) is an s-decomposition of / if it is a
maximal s-free tree of /.
Notice that the nite character of dependence implies, that unions of increasing
sequences of s-free trees of / are s-free trees of /. So for all s-saturated / there
is an s-decomposition of /.
We say that / is s-primary over an s-free tree (P, f, g) if / is s-primary over

/
t
[ t P.
5.8 Denition. Assume that (P, f, g) is an s-decomposition of /, / is s-
saturated. Let P = t
i
[ i < be an enumeration of P such that if t
i
t
j
then
i < j . Then we say that (/
i
)
i
is a generating sequence if the following holds
(i) for all i , /
i
/,
(ii) /
0
= ,
(iii) /
i+1
is s-primary over /
i
/
t
i
,
(iv) if 0 < i is limit then /
i
is s-primary over

j<i
/
j
.
5.9 Lemma. Assume that (P, f, g) is an s-free tree of /, / is s-saturated
and (/
i
)
i
is a generating sequence. Then for all 0 < i < , /
t
i

A
t

i
/
i
.
Proof. By Lemma 5.4 (i), it is enough to prove that for all i < , /
i
is
s-primary over
j<i
/
t
j
. We prove this by induction on i. In fact we need to
20
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


prove slightly more to keep the induction going: We show that /
i
is not only s-
constructible over
j<i
/
t
j
but that the natural construction works. Then the limit
cases are trivial and the successor cases follow from Lemma 5.4 (ii).
5.10 Denition. Assume / is s-saturated. We say that t(a, /) is a c-type
if for all s-saturated ( and B the following holds: If ( / is such that t(a, /) is
not orthogonal to ( and / a B, then there is b B / such that b
C
/.
Notice that the notion of c-type is a generalization of regular type.
5.11 Lemma. Assume M is superstable. Let / B be s-saturated and
/ ,= B. Then there is a singleton a B / such that t(a, /) is a c-type.
Proof. Since (M) = , by Lemma 1.1 it is easy to see that there is a singleton
a B / and nite A / such that the following holds: for all b B / and
B /, if there is an automorphism f of M such that f(a) = b and f(A) = B,
then t(b, /) does not split strongly over B (and so b
B
/). We show that a is as
wanted. Let s-saturated ( / be such that t(a, /) is not orthogonal to (. Since
B can now be any s-saturated model such that /a B, it is enough to show that
there is b B / such that b
C
/.
By Lemma 4.6, nd d such that d
C
/ and a ,
A
d. Let T be s-primary
over ( d. Then T
C
/ and a ,
A
T. For all i < , choose /
i
and a
i
so that
t(a
i
/
i
, T) = t(a /, T) and a
i
/
i

D
a /

j<i
(a
j
/
j
).
Claim. a / a
i
/
i
[ i < is indiscernible over ( and a / ,
C

i<
(a
i
/
i
).
Proof. The rst of the claims follow immediately from Corollary 3.5 (ii). For
a contradiction, assume that the second claim is not true. For all i < , we dene
B
i
as follows: We let B
0
be s-primary over / a and B
i+1
be s-primary over
B
i
/
i
a
i
. By Lemma 5.1, there is i < such that d
B
i
/
i
a
i
. Since
a / a
i
/
i
[ i < is indiscernible over ( and a /
C

i<
(a
i
/
i
),
/
i
a
i

C
a /

j<i
(a
j
/
j
). By Lemma 5.4 (ii), /
i
a
i

C
B
i
. But then
/
i
a
i

C
d, a contradiction. Claim.
By Claim and Corollary 3.5 (v), let n < be the least such that a / ,
C

i<n
(a
i
/
i
). Let /

be s-primary over / /
0

0<i<n
(/
i
a
i
). It is easy to
see that /
n

C
/

0<i<n
(/
i
a
i
). By Claim, /
0

C
/

0<i<n
(/
i
a
i
) and so
by Lemma 3.8 (iv) and the choice of n, / a
C
/
0

0<i<n
(/
i
a
i
) and so by
Lemma 3.6 and 3.2 (i), a
A
/
0

0<i<n
(/
i
a
i
). By Lemma 5.4 (i), a
A
/

.
Similarly we we see that a
0

A
0
/

. Then also a ,
A
a
1
.
By the choice of /
0
and a
0
there is f Aut(() such that f(/) = /
0
and
f(a) = a
0
. Let A
0
= f(A). By Corollary 3.5 (v) there is nite C /

such that
a ,
A
A
0
C a
0
. Choose B ( such that t(Aa, () does not split strongly over
B. Then there is g Saut(B) such that g(A
0
) (. Since a /
C
/

and every
h Aut(/

) belongs to Saut(B), we may assume that

() a /
C
g(C) A
0
C.
21
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


Then t(g(A
0
C), Aa) = t(A
0
C, Aa). Choose h Saut(Ag(A
0
)) such that
h(g(C)) /. By (*), t(a, / g(C)) does not split strongly over A and so it does
not split strongly over Ag(A
0
). Then t(g(A
0
) h(g(C)), Aa) = t(A
0
C, Aa).
Choose b B such that t(g(A
0
)h(g(C))b, Aa) = t(A
0
Ca
0
, Aa). Then by
Corollary 3.5 (v) and the choice of C, a ,
A
b and so by Lemma 3.2 (iii), b B/
(b is a singleton). By the choice of A, t(b, /) does not split strongly over g(A
0
).
By Lemma 3.2 (iii), b
C
/.
5.12 Denition.
(i) We say that M has s-SP (structure property) if every s-saturated / is
s-primary over any s-decomposition of /.
(ii) Let (M). We say that M has -dop if there are F
M

-saturated /
i
,
i < 4, and a , /
3
such that
(a) /
0
/
1
/
2
,
(b) /
1

A
0
/
2
,
(c) /
3
is F
M

-primary over /
1
/
2
,
(d) t(a, /
3
) is orthogonal to /
1
and to /
2
.
We say that M has -ndop if it does not have -dop.
5.13 Theorem. Assume M is superstable and has (M)-ndop. Then M
has s-SP.
Proof. Let / be s-saturated and (P, f, g) an s-decomposition of /. Let
(/
i
)
i
be a generating sequence and P = t
i
[ i < be the enumeration of P
from the denition of a generating sequence.
Claim: /

= /.
Proof. Assume not. For all a / /

let i
a
be the least ordinal such that
t(a, /

) is not orthogonal to /
i
a
. Let a //

be any sequence such that

(i) for some l either t(a, /
l
) is a c-type and a
A
l
/

or t(a, /
t
l
) is a
c-type and a
A
t
l
/

and
(ii) among these a, i = i
a
is the least.
By Lemma 5.11 there is at least one such a.
There are two cases:
Case 1: For some l < t(a, /
t
l
) is a c-type and a
A
t
l
/

. Let t

t
l
be
the least t such that t(a, /
t
l
) is not orthogonal to /
t
. Since t(a, /
t
l
) is a c-type
choose b so that
(1) b
A
t

/
t
l
and
(2) b /
t
l
[a] /
t
l
, where /
t
l
[a] / is s-primary over /
t
l
a.
Then if (t

exists, by (2) and Lemmas 4.6 and 5.4 (i), t(b, /

t
l
) is orthogonal to
/
(t

)
and so by (1) and Lemma 4.6 it is easy to see that t(b, /
t
) is orthogonal to
/
(t

)
.
By (1), (2) and Lemma 5.4 (i), b
A
t

.
We dene ((P

), f

, g

) as follows:
22
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


(i) P

= P t, t a new node,
(ii) for all u P , u

t i u _ t

(iii) f

P = f and f

(t) = b,
(iv) g

P = g and g

(t) / is s-primary over A

t
b.
Subclaim. ((P

), f

, g

) is an s-free tree of /.
Proof. (i), (ii), (iii) and (iv) in the Denition 5.6 are clear. So we prove (v):
Let T P

, u P

and V P

be as in Denition 5.6 (v). There are four

cases:
Case a: t T V . Let T

= T t and /
T
/

be s-primary over
/
d
[ d T

. By the choice of b and Lemma 5.4 (i),

A
t

A
t

/
T

vV
/
v
.
By Lemmas 3.2 (i) and 3.6,

vV
/
v

A
T

/
t
.
By Corollary 3.5 (iv), the assumption that (P, f, g) is s-free tree of / and Lemma
5.4 (i),

vV
/
v

A
u
/
T
/
t
.
By Lemma 3.6,

dT
/
d

A
u

vV
/
v
.
Case b: t V T : Exactly as the Case a.
Case c: t V T : Because t T P , u t. Since t V , u = t. Then
because u , P ,
dT
/
d
= /
u
, and the claim follows from Lemma 3.2 (iv).
Case d: t , T V : Immediate by the assumption that (P, f, g) is an s-free
tree of /.
Subclaim.
Subclaim contradicts the maximality of P . So Case 1 is impossible and we are
in the Case 2:
Case 2: l is such that t(a, /
l
) is a c-type and a
A
l
/

. Let B /
be s-primary over /

a. Clearly i(= i
a
) l and so let b

be the element given

by t(a, /
l
) being a c-type: b

A
i
/
l
and b

/
l
[a] /
l
, where /
l
[a] B is
s-primary over /
l
a. By Lemma 5.11 we may choose b so that t(b, /
i
) is a c-type
and b /
i
[b

] /
i
, where /
i
[b

] B is s-primary over /
i
b

. Then b
A
i
/

,
b , /
i
and i
b
i(= i
a
).
1. i is not a limit > 0. This is because otherwise by Lemma 5.5 (ii), t(b, /
i
)
is not orthogonal to /
j
for some j < i. Then t(b, /

) is not orthogonal to /
j
i.e.
i
b
< i
a
. This contradicts the choice of a.
2. i is not a successor > 1. Assume it is, i = j +1. Then /
i
is s-primary over
/
j
/
t
j
and by Lemma 5.9, /
j

A
t

j
/
t
j
. (Notice that since Case 1 is not possible,
A
j+1
,= /
t
j
.) By the choice of a t(b, /
i
) is orthogonal to /
j
. So by (M)-ndop
23
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


t(b, /
i
) is not orthogonal to /
t
j
. Then as in Case 1 we get a contradiction with
the maximality of (P, f, g). Alternatively, we can nd c such that it satises the
assumptions of Case 1, which is a contradiction.
3. i is not 0 or 1. Immediate, since Case 1 is not possible.
Clearly 1 and 2 above contradict 3. So also Case 2 imply a contradiction.
Claim.
Let ( / be F
M

-primary over

/
t
[ t P. We want to show that ( = /.
For this we choose a generating sequence (/
i
)
i
, so that /
i
B for all i . By
the claim above /

= / and so ( = /.
6. On nonstructure
6.1 Denition. We say that M has -sdop if the following holds: there are
F
M

-saturated /
i
, i < 4, and I = a
i
[ i < (M), a
i
/
3
, such that
(a) /
0
/
1
/
2
, /
3
is F
M

-primary over /
1
/
2
,
(b) /
1

A
0
/
2
,
(c) I is an indiscernible sequence over /
1
/
2
and if i < j < (M) then
a
i
,= a
j
.
As in [Hy], we can prove non-structure theorems from -sdop. (In [Sh2], this
was the formulation of dop, which was used to get non-structure.)
In this chapter we show that dop and sdop are essentially equivalent i.e. (M)
+
-
sdop implies (M)
+
-dop and (M)-dop implies
r
(M)
+
-sdop, where
r
(M) is the
least regular cardinal (M).
6.2 Lemma. Assume M is -stable and =
+
. If M has -sdop then it
has -dop.
Proof. Let I and /
i
, i < 4, be as in the denition of -sdop. We need to
show that there is M-consistent type p over /
3
such that (d) in Denition 5.12 (ii)
is satised. We show that Av(I, /
3
) is the required type.
By Lemma 2.4 (iii), let a be such that t(a, /
3
) = Av(I, /
3
). For a contradic-
tion, by Lemma 4.6, let b be such that
(i) a ,
A
3
b,
(ii) b
A
1
/
3
.
Let (
i
/
i
, i < 4 be F
M

-saturated models of cardinality such that

(1) (
i
/
i
, (
1
(
2
= (
0
, (
3
/
1
= (
1
, (
3
/
2
= (
2
and I (
3
,
(2) a b
C
3
/
3
and a ,
C
3
b,
(3) a b (
3

C
1
/
1
and a b (
3

C
2
/
2
,
(4) for all c (
3
there is D (
1
(
2
of power , such that t(c, D) F
M

-isolated
t(c, /
1
/
2
).
We can see the existence of the sets as in the proof of Theorem 3.14 (the only
non-trivial part being to guarantee that the models are F
M

-saturated).
Let a

/
3
be such that it realizes Av(I, (
3
).
Claim. t(a

, (
3
) F
M

-isolates t(a

, (
3
/
1
/
2
).
24
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


Proof. Assume not. Then there is d (
3
such that t(a

d, (
1
(
2
) does not
F
M

-isolate t(a

d, /
1
/
2
).
Subclaim. There is a

I such that t(a

d, (
1
(
2
) = t(a

d, (
1
(
2
).
Proof. By Lemma 1.2 (v), there is i < (M) such that t(d, (
1
(
2
I) does
not split over (
1
(
2
a
j
[ j < i. Since I is indiscernible over C
1
(
2
, a

= a
i
is
as wanted. Subclaim.
Clearly Subclaim contradicts (4) above. Claim.
Choose b

/
1
so that t(b

, (
1
) = t(b, (
1
). By (3), t(b

, (
3
) = t(b, (
3
). By
Claim, a

C
3
b

. Let f be an automorphism such that f(b) = b

and f (
3
= id
C
3
.
Then f(a) contradicts Claim.
6.3 Theorem. Let
r
(M) be such that M is -stable. Then
r
(M)-dop
implies
+
-sdop.
Proof. Let /
i
, i < 4, and p S(/
3
) be as in the the denition of
r
(M)-dop.
By Lemma 4.5, as in the proof of Lemma 3.14, we nd these so that [/
3
[
r
(M).
Let B
0
/
0
be F
M

+
-saturated such that B
0

A
0
/
3
. Let B
i
, i 1, 2 be s-
primary over /
i
B
0
. Let B
3
be s-primary over B
1
B
2
/
3
. Let (
i
, i 1, 2,
be F
M

+
-primary over B
i
such that (
1

B
1
B
3
and (
2

B
2
B
3
(
1
.
Let q S(/
3
) be any type such that it is orthogonal to /
1
and /
2
. Our rst
goal is to show that there is only one q

S((
1
(
2
B
3
) which extends q .
Claim 1. B
1
/
2
is F
M

r
(M)
-constructible over /
1
B
0
/
2
and for all
b B
1
, there is B /
1
B
0
of power <
r
(M) such that t(b, B) F
M

r
(M)
-isolates
t(b, B
0
/
1
/
2
).
Proof. Follows immediately from the proof of Lemma 5.4 (ii) and Theorem 5.3
(iii). Claim 1.
Claim 2 B
1
B
2
is F
M

r
(M)
-constructible over B
1
B
0
/
2
and for all b
B
2
, there is B /
2
B
0
of power <
r
(M) such that t(b, B) F
M

r
(M)
-isolates
t(b, B
1
/
2
).
Proof. As Claim 1. Claim 2.
Claim 3 B
1
B
2
/
3
is F
M

r
(M)
-constructible over /
3
B
0
.
Proof. By Claims 1 and 2, B
1
B
2
is F
M

r
(M)
-constructible over /
1
/
2
B
0
. So
it is enough to show that for all a /
3
, t(a, /
1
/
2
) F
M

r
(M)
+
-isolates t(a, B
1
B
2
).
Assume not. Choose b
1
B
1
and b
2
B
2
so that t(a, /
1
/
2
) does not
F
M

r
(M)
+
-isolate t(a, /
1
/
2
b
1
b
2
). Choose A
1
/
1
, A
2
/
2
and B
0
B
0
of power <
r
(M) such that
(i) t(a, A
1
A
2
) F
M

r
(M)
-isolates t(a, /
1
/
2
),
(ii) t(b
1
, A
1
B
0
) F
M

r
(M)
-isolates t(a, /
1
/
2
B
0
) and t(b
2
, A
2
B
0
) F
M

r
(M)
-
isolates t(b
2
, B
1
/
2
B
0
),
(iii) A
1
/
0
= A
2
/
0
= A
0
and for all c A
1
A
2
, t(c, B
0
) does not split
strongly over A
0
.
By Lemma 1.9 (v), choose f Aut(A
0
) so that f(B
0
) /
0
and for all c B
0
,
f(c) E
m
min,A
0
c. Let B

0
= f(B
0
). Then by (iii), t(B

0
, A
1
A
2
) = t(B
0
, A
1
A
2
).
25
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


Choose b

i
/
i
so that t(b

i
B

0
, A
i
) = t(b
i
B
0
, A
i
), i 1, 2. By (ii) t(b

1
b

0
, A
i
) = t(b
1
b
2
B
0
, A
i
). Clearly this contradicts (i). Claim 3.
Claim 4. B
3
is F
M

r
(M)
-primary over /
3
B
0
.
Proof. Immediate by Claim 3 and the choice of B
3
. Claim 4.
By Claim 4 and Lemma 5.4, there is exactly one q

S(B
3
) such that q q

.
By Corollary 4.8, q

is orthogonal to B
1
and B
2
. So if a realizes q

, then a
B
3
(
1
.
Then by Corollary 5.5 (i), there is exactly one q

S((
1
(
2
B
3
), which extends
q .
Now choose a
i
, i <
r
(M), so that for all i, t(a
i
, /
3
) = p and a
i

A
3

j<i
a
j
.
Then I = a
i
[ i <
r
(M) is indiscernible over /
3
and by Corollary 5.5 (iii), for
all n < , t(a
0
... a
n
, /
3
) is orthogonal to /
1
and /
2
. So, by what we showed
above, I is indiscernible over (
1
(
2
B
3
and for all i <
r
(M), t(a
i
, /
3

j<i
a
j
)
F
M

+
-isolates t(a
i
, (
1
(
2
B
3

j<i
a
j
). So there is an F
M

+
-primary model (
3
over
(
1
(
2
B
3
such that I (
3
.
So to get
+
-sdop, it is enough to show that (
3
is F
M

+
-primary over (
1
(
2
.
By Claim 3 and the choice of B
3
, B
3
is F
M

r
(M)
-constructible over B
1
B
2
.
Claim 5. For all c (
1
there is B B
1
of power such that t(c, B)
F
M

+
-isolates t(c, B
3
).
Proof. Assume not. Choose B
1
B
1
of power and c (
1
so that
(i) t(c, B
1
) F
M

+
-isolates t(c, B
1
),
(ii) t(c, B
1
) does not F
M

+
-isolate t(c, B
3
).
By (ii) above, choose b B
3
, B
0
B
0
C
1
B
1
and C
2
B
2
(iii) [C
1
C
2
[ <
r
(M),
(iv) t(b, C
1
C
2
) F
M

r
(M)
-isolates t(b, B
1
B
2
),
(v) t(c, B
1
) does not F
M

+
-isolate t(c, B
1
C
1
C
2
b),
(vi) for all a B
1
C
1
, t(a, B
2
) does not split strongly over B
0
and [B
0
[ .
Since B
0
is F
M

+
-saturated and M is -stable, we can nd f Aut(B
0
) such that
f(C
2
) B
0
and for all a C
2
, f(a) E
m
min,B

0
a. Then by (vi), t(f(C
2
), B
0
B
1

C
1
) = t(C
2
, B
0
B
1
C
1
). Choose b

B
1
so that t(b

f(C
2
), C
1
) = t(b C
2
, C
1
).
Then by (iv), t(b

f(C
2
), B
0
B
1
C
1
) = t(b C
2
, B
0
B
1
C
1
). By (vi) t(c, B
1
)
does not F
M

+
-isolate t(c, B
1
(
1
f(C
2
) b

). Clearly this contradicts (i). Claim

5
So B
3
(
1
is F
M

r
(M)
-constructible over (
1
B
2
.
Claim 6. For all c (
2
there is B B
2
of power such that t(c, B)
F
M

+
-isolates t(c, (
1
B
3
).
Proof. As Claim 5 above. Claim 6.
So B
3
(
1
C
2
is F
M

r
(M)
-constructible (and so F
M

+
-constructible) over (
1
(
2
.
By the choice of (
3
, this implies that (
3
is F
M

+
-primary over (
1
(
2
.
Notice that in Theorem 6.3 the assumption, M is -stable, is not necessary.
We can avoid the use of it by Lemma 3.15.
6.4 Lemma. Assume > (M). Then -dop implies -dop.
26
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


Proof. Let /
i
, i < 4, and a as in the denition of -dop. Choose F
M

-
saturated B
0
/
0
such that B
0

A
0
/
1
/
2
. Let B
1
be F
M

-primary over
B
0
/
1
, B
2
be F
M

-primary over B
0
/
2
and B
3
be F
M

-primary over B
1
B
2
.
Clearly we can choose the sets so that /
3
B
3
and a
A
3
B
3
. By Lemmas 5.4 (iii)
and 3.8 (iv), B
1

B
0
/
2
. Then /
2

A
0
B
1
and so /
2

A
1
B
1
. By Lemma 5.4 (iii),
(1) /
3

A
1
B
1
.
Similarly,
(2) /
3

A
2
B
2
.
Also by Lemmas 5.4 (iii) and 3.8 (iv), B
1

B
0
B
2
.
By (1), (2), Lemma 4.5 and Corollary 4.8, t(a, B
3
) is orthogonal to B
1
and to
B
2
.
6.5 Corollary. (M)-dop implies
r
(M)
+
-sdop.
Proof. Immediate by Lemma 6.4 and Theorem 6.3.
We nish this paper by giving open problems:
6.6 Question. What are the relationships among the following properties:
(1) a
A
A,
(2) a ,
A
a,
(3) t(a, A) is unbounded?
Notice that (1) does not imply (2) nor (3) (fails already in the classical case),
(3) implies (2) (Lemma 3.2 (v)) and (1)(2) implies (3) (just choose a
i
, i < [M[ ,
so that t(a
i
, A) = t(a, A) and a
i

A

j<i
a
j
).
6.7 Question. Does Corollary 4.8 hold without the assumption that the sets
are strongly F
M

r
(M)
-saturated?
6.8 Question. Does the following hold: If M is superstable, then for all A
there exists an a-primary set over A?
27
6
2
9


r
e
v
i
s
i
o
n
:
1
9
9
8
-
1
1
-
2
4







m
o
d
i
f
i
e
d
:
1
9
9
9
-
1
0
-
0
4


References.
[Hy] T. Hyttinen, On nonstructure of elementary submodels of a stable homogeneous
structure, Fundamenta Mathematicae, vol. 156, 1998, 167-182.
[Sh1] S. Shelah, Finite diagrams stable in power, Annals of Mathematical Logic, vol.
2, 1970, 69-118.
[Sh2] S. Shelah, Classication Theory, Stud. Logic Found. Math. 92, North-Holland,
Amsterdam, 2nd rev. ed., 1990.
Tapani Hyttinen
Department of Mathematics
P.O. Box 4
00014 University of Helsinki
Finland
Saharon Shelah
Institute of Mathematics
The Hebrew University
Jerusalem
Israel
Rutgers University
Hill Ctr-Bush
New Brunswick
New Jersey 08903
U.S.A.
28