Saharon Shelah - Characterizing An - Epsilon Saturated Model of Superstable NDOP Theories by Its L, - Epsilon Theory

(
4
0
1
)

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

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

CHARACTERIZING AN
-SATURATED
MODEL OF SUPERSTABLE NDOP
THEORIES BY ITS L
,
-THEORY
SH401
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
MSRI
Berkeley, CA USA
Abstract. Assume a complete countable rst order theory is superstable with
NDOP. We knew that any
-saturated model of the theory is
-prime over a
non-forking tree of small models and its isomorphism type can be characterized by
it L
,
(dimension quantiers)-theory; or if you prefer - appropriate cardinal invari-
ants. We go here one step further providing cardinal invariants which are as nitary
as seems reasonable.
Partially supported by the United States-Israel Binational Science Foundation and the NSF, and
I thank Alice Leonhardt for the beautiful typing.
Revised with proofreading for the Journal
Done Oct/89
Publ No. 401
Latest Revision - 05/Dec/20
Typeset by A
M
S-T
E
X
1
(
4
0
1
)

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

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

2 SAHARON SHELAH
Introduction
After the main gap theorem was proved (see [Sh:c]), in discussion, Harrington
expressed a desire for a ner structure - of nitary character (when we have a struc-
ture theorem at all). I point out that the logic L
,
0
(d.q.) (where d.q. stands for
dimension quantier) does not suce: suppose; e.g. for T = Th(
2, E
n
)
n<
where (, )E
n
(, ) =: n = n and for S

2 dene M
S
= M (, ) :
[ S <
1
] and [

2S < ]. Hence, it seems to me we should
try L
,
(d.q.) (essentially, in C we can quantify over sets which are included in

the algebraic closure of nite sets, see below 1.1, 1.3), and Harrington accepts this
interpretation. Here the conjecture is proved for
-saturated models.
I.e., the main theorem is M
L
,
(d.q.)
N M

= N for
-saturated models
of a superstable countable (rst order) theory T without dop. For this we analyze
further regular types, dene a kind of innitary logic (more exactly, a kind of type
of a in M), looking only up in the denition (when thinking of the decomposition
theorem). Recall that for as
-saturated model M of a superstable DNOP theory

an
-decomposition is M
, a
: T ) where
(a) I
>
ord is nonempty closed under initial segments
(b) M
M is
-saturated
(c) I M
(d) if = ) I then M
is
-prime over M
and tp(a
, M
) is
orthogonal to M
for and (the last is not essential but claries)

(e) M
: I) is non-forking enough: for every I the set a
:
Suc
I
() M is independent over M
.
The point is that if = ), M
, a
are chosen then to a large extent M
, a
:
I) is determined. But the amount of to a large extent which suces in
[Sh:c], is not sucient here, we need to nd a ner understanding. In particular,
we certainly do not like to know (M
, a
). So we consider a pair (A, B) where

A M
, A a
B M
, stp
(B, A) stp
(B, M
) and we try to dene the

type of such pairs in a way satisfying:
(a) it can be impressed in our logic L
,
(b) it expresses the essential information in M
, a
: I).
To carry out the isomorphism proof we need: (1.27) the type of the sum is the sum
of types (innitary types) assuming rst order independence. The main point of the
proof is to construct an isomorphism between M
1
and M
2
when M
1

L
,
(d.q.)
M
2
, Th(M
) = T where T and
L
,
(q.d.)
are as above. So by [Sh:c, X] it is
(
4
0
1
)

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

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

CHARACTERIZING MODELS OF NDOP THEORIES 3
enough to construct isomorphic decompositions. The construction of isomorphic
decompositions is by approximations, in stage n, n levels of the decomposition
tree are approximated, i.e. we have I
n

n
Ord and a
n,
for I
n
, = 1, 2
such that tp( a
n,1
0
a
n,1
1
. . . a
n,1
, , M) = tp( a
n,2
0
a
n,2
1
. . . a
n,2
, , M
2
) with a
,
being -nite, so in stage n + 1, choosing a

n+1,
we cannot take care of all types

a
n+1,
a
,
so addition theorem takes care. So though we are thinking on
-
decompositions (i.e., the M
s are
-saturated), we get just a decomposition.

In the end of 1 (in 1.37) we point up that the addition theorem holds in fuller
generalization. In the second section we deal with a ner type needed for shallow
T, in the appendix we discuss how absolute is the isomorphism type.
Of course, we may consider replacing
-saturated models of an NDOP su-

perstable countable T by models of an NDOP
0
-stable countable T. But the
use of -nite sets seems considerably less justiable in this context, it seems more
reasonable to use nite sets, i.e., L
,
0
(d.q.). But subsequently Hrushovski and
Bouscaren proved that even if T is
0
-stable, L
,
0
(d.q.) is not sucient to char-
acterize models of T up to isomorphism. This is not sucient even if one considers
the class of all
-saturated models rather than all models. The rst example is
0
-stable shallow of depth 3, and the second one is superstable (non
0
-stable),
NOTOP, non-multi-dimensional.
If we deal with
-saturated models of shallow (superstable NDOP) theories T,

we can bound the depth of the quantication = DP(T); i.e. L
suce.
We assume the reader has a reasonable knowledge of [Sh:c, V,1,2] and mainly
[Sh:c, V,3] and of [Sh:c, X].
Here is a slightly more detailed guide to the paper. In 1.1 we dene the logic
L
,
and in 1.3 give a back and forth characterization of equivalence in this logic
which is the operative denition for this paper.
The major tools are dened in 1.7, 1.11. In particular, the notion of tp
dened
in 1.5 is a kind of a depth look-ahead type which is actually used in the nal
construction. In 1.28 we point out that equivalence in the logic L
,
implies
equivalence with respect to tp
for all . Proposition 1.14 contains a number of

important concrete assertions which are established by means of Facts 1.16-1.23.
In general these explain the properties of decompositions over a pair
_
B
A
_
. Claim
1.27 (which follows from 1.26) is a key step in the nal induction. Denition 1.30
establishes the framework for the proof that two
-saturated structures that have

the same tp
are isomorphic. The induction step is carried out in 1.35.

I thank Baldwin for reading the typescript pointing needed corrections and writ-
ing down some explanations.
-1.1 Notation: The notation is of [Sh:c], with the following additions (or reminders).
If = ) then we let
= ; for I a set of sequences ordinals we let

(
4
0
1
)

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

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

4 SAHARON SHELAH
Suc
T
() = : for some , = ) I.
We work in C
eq
and for simplicity every rst order formula is equivalent to a relation.
(1) means orthogonal (so q is p means q is orthogonal to p),
remember p A means p orthogonal to A; i.e. p q for every
q S(ac(A)) (in C
eq
)
(2)
a
means almost orthogonal
(3)
w
means weakly orthogonal
(4)
a
B
and a/B means tp( a, B)
(5)
A
B
or A/B means tp
(A, B)
(6) A + B means A B
(7)
A
B
i
: i < means B
i
: i < is independent over A
(8) A
B
C means A, C is independent over B
(9) C
i
: i < is independent over (B, A) means that
1
j < tp
(C
j
, C
i
B : i ,= j) does not fork over A
(10) regular type means stationary regular type p S(A) for some A
(11) for p S(A) regular and C a set of elements realizing p, dim(C, p) is
Max[I[ : I C is independent over A
(12) ac(A) = c : tp(c, A) is algebraic
(13) dc(A) = c : tp(c, A) is realized by one and only one element
(14) Dp(p) is depth (of a stationary type, see [Sh:c, X,Denition 4.3,p.528,Denition
4.4,p.529]
(15) Cb(p) is the canonical base of a stationary type p (see [Sh:c, III,6.10,p.134])
(16) B is
-atomic over A if for every nite sequence

b from A, for some nd
A
0
A we have stp(
b, A
0
)) stp(
b, A), equivalently for some -nite

A
0
acl(A) we have tp(
b, A
0
)) tp(
b, acl(A)).
1
Actually by the non-forking calculus this is equivalent to {C
i
: i } is independent over A
where we let C
= B.
(
4
0
1
)

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

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

1
-saturated models
We rst dene our logic, but as said in 0, we shall only use the condition from
1.4. T is always superstable complete rst order theory.
1.1 Denition. 1) The logic L
,
is slightly stronger than L

,
0
, it consists of
the set of formulas in L
,|T|
+ such that any subformula of of the form ( x) is
actually the form
( x
0
, x
1
)
_
_
1
( x
1
, y) &
i<g x
1
_
i
(x
1
i
, x
0
) & (
<
0
z)
i
(z, x
0
)
_
_
_
,
with x
0
nite, x
1
not necessarily nite but of length < [T[
+
; so says x
1
ac( x
0
); note that always our nal proof of the theorem uses [T[
0
.
2) L
,
(d.q.) is like L
,
but we have cardinality quantiers and moreover di-

mensional quantiers (as in [Sh:c, XIII,1.2,p.624]), see below.
3) The logic L
consist of the formulas of L

,
such that has quantier depth

< (but we start the inductive denition by dening the quantier depth of all
rst order as zero).
4) L
(d.q.) is like L
but we have cardinality quantiers and moreover di-

mensional quantiers.
1.2 Remark. 1) In fact the dimension quantier is used in a very restricted way
(see Denition 1.9 and Claim 1.28 + Claim 1.30).
2) The reader may ignore this logic altogether and use just the characterization of
equivalence in claim 1.4.
1.3 Convention. 1) T is a xed rst order complete theory, C is the monster
model, as in [Sh:c], so is -saturated; C
eq
is as in [Sh:c, III,6.2,p.131]. We work in
C
eq
so M, N vary on elementary submodels of C
eq
of cardinality < . We assume T
is superstable with NDOP (countability is used only in the Proof of 1.5 for book-
keeping, i.e., in the proof of 1.30 (and 1.29).
Remember a, b, c, d denote members of C
eq
, a,
b, c,

d denote nite sequences of mem-
bers of C
eq
, A, B, C denote subsets of C
eq
of cardinality < .
Remember ac(A) is the algebraic closure of A, i.e.
b : for some rst order and n < , (x, y) and a A we have C
eq
[= [b, a] &
(
n
y)(y, a)
and a denotes Rang( a) in places where it stands for a set (as in ac( a). We write
(
4
0
1
)

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

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

6 SAHARON SHELAH
a A instead of a
>
(A).
2) A is -nite, if for some a
>
A, A = ac( a). (So for stable theories a subset
of an -nite set is not necessarily -nite but as T is superstable, a subset of an
-nite set is -nite as if B ac( a),
b B is such that tp( a, B) does not fork over
b, then trivially ac(
b) A and if ac(
b) ,= B, tp
(B, a
b) forks over B, hence

([Sh:c, III,0.1]) tp( a, B) forks over

b, a contradiction.
So if ac(A) = ac(B), then A is -nite i B is -nite).
3) When T is superstable by [Sh:c, IV,Table 1,p.169] for F = F
a
0
, all the axioms
there hold and we write
instead of F and may use implicitly the consequences

in [Sh:c, IV,3].
We may instead Denition 1.1 use directly the standard characterization from 1.4;
as actually less is used we state the condition we shall actually use:
1.4 Claim. For models M
1
, M
2
of T we have M
1

L
,
(d.q.)
M
2
i
there is a non-empty family F such that:

(a) each f F is an (M
1
, M
2
)-elementary mapping, (so Dom(f) M
1
,
Rang(f) M
2
)
(b) for f F, Dom(f) is -nite (see 1.3(2)) above)
(c) if f F, a
( = 1, 2) then for some g F we have:

f g and ac( a
1
) Dom(f) and ac( a
2
) Rang(f)
(d) if f a
1
, a
2
) F and tp(a
1
, Dom(f)) is stationary and regular
then dim(a
1
1
M
1
: f a
1
1
, a
2
) F, M
1
)
= dim(a
1
2
M
2
: f a
1
, a
1
2
) F, M
2
).
Our main theorem is
1.5 Theorem. Suppose T is countable (superstable complete rst order theory)
with NDOP.
Then
(1) the L
,
(d.q.) theory of an
-saturated model characterizes it up to iso-

morphism.
(2) Moreover, if M
1
, M
2
are
-saturated models of T (so M
C
eq
) and
M
0
,M
1
of 1.4 holds, then M
1
, M
2
are isomorphic.

By 1.4, it suces to prove part (2).
(
4
0
1
)

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

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

The proof is broken into a series of claims (some of them do not use NDOP, almost
all do not use countability; but we assume T is superstable complete all the time
(1.7(1)).
1.6 Discussion: Let us motivate the notation and Denition below.
Recall from the introduction that we are thinking of a triple (M, N, a) which
may appear in
-decomposition M
, a
: I) of N, in the sense that for some

I<> we have (M, M
, a) = (M
, M
, a
) so M, M
are
-saturated, a
, M
is
-prime over M + a and tp(a, M) is regular. But this is too large

for us hence we consider an approximation (A, B) where A M(= M
), A
B M
(= M
)), a = a
B and B/M(= B/M
) does not fork over A. We

would like to dene the -type of (A, B) in N, which tries to say something on the
decomposition above (M, M
, a) = (M
, M
, a
), i.e., on M
, a
: I). There
are two natural successor of (A, B) we may choose in this context: the rst 1.7
below replaces (A, B) to (A
, B
) such that A A
M(= M
), B B
(=
M
) and (as M
is
-prime over M + a) we have stp
(B
, A
B) stp(B
, M),
so tp(B
, A
B) is almost orthogonal to A
; we can think of this as advancing in

the same model; in other words as A, B are -nite, we have to increase them in
order to capture even (M, M
). This is formalized by
a
in Denition 1.7 below.
The second is to pass from (M
, M
, a) to (M
, M
, a
) for some an immediate

successor (in I) of I. So the old B is included in the new A
and B
= A
a
where tp(a, A
) is regular and is orthogonal to A (as in the decomposition we require

tp(a
, M
)(M
when
). This is formalized by
b
in Denition 1.7 below.
1.7 Denition. 1) = (A, B) : A B are -nite. Let
(M) = (A, B) : A B M.
2) For members (A, B) of we may also write
_
B
A
_
; if A B we mean
_
BA
A
_
.
3)
_
B
1
A
1
_

a
_
B
2
A
2
_
(usually we omit a) if (both are in and)
A
1
A
2
, B
1
B
2
, B
1
A
1
A
2
and
B
2
B
1
+A
2
a
A
2
.
4)
_
B
1
A
1
_

b
_
B
2
A
2
_
if A
2
= B
1
, B
2
A
2
=

b and

b
A
2
is regular orthogonal to A
1
.
5)
is the transitive closure of

a

b
. (So it is a partial order, whereas in
general
a

b
and
b
are not).
6) We can replace A, B by sequences listing them (we do not always strictly distin-
guish).
Remark. The following observation may clarify.
1.8 Observation. If
_
B
1
A
1
_

_
B
2
A
2
_
then we can nd B
: n) and
(
4
0
1
)

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

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

8 SAHARON SHELAH
c
: 1 < n) for some n 1, satisfying

_
B
1
A
1
_

b
_
B
1
B
0
_
, c
+1
,
c
regular,
B
+1
c
+B
a
B
, A
2
= B
n1
, B
2
= B
n
.
Remark. 1) Note that actually
a
is transitive. This means that in a sense
b
is
enough,
a
inessential.
2) We may in 1.7(4) use

b = c), does not matter.
Proof. By the denition of
there are k < and

_
B
_
for k such that:
_
B
_

x()
_
B
+1
A
+1
_
for k and x() a,b and
_
B
0
A
0
_
=
_
B
1
A
1
_
,
_
B
k
A
k
_
=
_
B
2
A
2
_
and
without loss of generality x(2) = a,x(2 + 1) = b. Let N
0
C be
-prime
over such that A
0
N
0
, B
0
A
0
N
0
and f
0
= id
A
0
. We choose by induction on
k, N
+1
, f
+1
such that:
(a) Dom(f
+1
) = B
(b) N
N
+1
(c) if x() = b then f
+1
is an extension of f
(which necessarily has domain

A
, check) with domain B
such that f
(B
(A
)
N
and N
+1
is
-prime
over N
(B
)
(d) if x() = a, then f
+1
maps A
into N
1
, B
into N
and N
+1
= N
.
This is straightforward. Now on N
: k + 1) we repeat the argument (of

choosing B
: n)) in the proof of 1.14(6) above, i.e., choose B
by
downward induction on large enough as required.
1.8
1.9 Denition. 1) We dene tp
[
_
B
A
_
, M] (for A B M, A and B are
-nite and is an ordinal) and S
(
_
B
A
_
, M), S
(A, M) and S
r
(
_
B
A
_
, M), S
r
(A, M)
by induction on (we mean simultaneously; of course, we use appropriate vari-
ables):
(a) tp
0
[
_
B
A
_
, M] is the rst order type of A B
(b) tp
+1
[
_
B
A
_
, M] = the triple Y
1,
A,B,M
, Y
2,
A,B,M
, tp
(
_
B
A
_
, M))
where:
Y
1,
A,B,M
=:
_
tp
[
_
B
_
, M] : for some A
, B
we have
_
B
A
_

a
_
B
_
(M)
_
,
(
4
0
1
)

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

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

and Y
2,
A,B,M
=:
_
,
M,B
) : S
r
(B, M)
_
where
M,B
= dim
__
d : tp
[
_
B+d
B
_
, M] =
_
, B
_
:
(c) for a limit ordinal, tp
[
_
B
A
_
, M] = tp
[
_
B
A
_
, M] : < )
(this includes = , really |M|
+
suce).
(d) S
(A, M) =
_
tp
[
_
B
A
_
, M] : for some B such that B M,
and
_
B
A
_
(M)
_
(e) S
r
(
_
B
A
_
, M) =
_
tp
[
_
B+c
B
_
, M] : for some c M we have
c
B
A and
c
B
is regular
_
(f) S
r
(A, M) =
_
tp
[
_
A+c
A
_
, M] : c M and
c
A
regular
_
.
2) We dene also tp
[A, M], for A an -nite subset of M:

(a) tp
0
[A, M] = rst order type of A
(b) tp
+1
[A, M] is the triple Y
1,
A,M
, Y
2,
A,M
, tp
[A, M]) where

Y
1,
A,M
=: S
(A; M) and
Y
2,
A,M
=:
_
, dimd M : tp
[
_
A+d
A
_
, M] = ) : S
r
(A, M)
_
(c) tp
[A, M] = tp
(A, M) : < )
3) tp
[M] = tp
[, M].
1.10 Discussion: Clearly tp[
_
B
A
_
, M] is intended, on the one hand, to be expressible
by our logic and, on the other hand, to express the isomorphism type of M in the
direction of
_
B
A
_
. To really say it we need to go back to the
-decompositions of
M, a central notion of [Sh:c, Ch.X].
For the readers benet, by the referee request, let us review informally the
proof in [Sh:c, Ch.X]. Let M be an
-saturated model, and we choose M
:
I
n
Ord), a
: I
n+1
) by induction on n. For n = 0, of course,
I
0
Ord = <>, we let N
<>
M be
-prime over and let I

<>
a maximal
subset of c M : tp(c, N
<>
) regular which is independent over N
<>
, let a
<>
:
< [I
<>
[) list I
<>
. Similarly for n + 1, I
n+1
Ord let N
M be
-
prime over M
+ a
, let I
be a maximal subset of c M : tp(c, M
) is regular
(
4
0
1
)

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

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

10 SAHARON SHELAH
orthogonal to M
independent over N
. Lastly, let c
<>
: < [I
[) list I
and let I
n+1
Ord = < >: I
n
Ord and < [I
[.
To carry this we use the existence of
-prime models (and the local character

of indpendent). Also looking at the set M
: I, its rst order type is

determined by the non-forking calculus. In fact, for any I<>, the set
N
: I, N
: ( ) and I are independent over N
. Let N M
be
-prime over N
: I, now if M = N we are done decomposing M, if not

some c MN realize a regular type (we use density of regular types). By NDOP,
the tp(c, N) is not orthogonal to some N
. Choose of minimal length hence

tp(c, M
) N
. By properties of regular types, without loss of generality tp(c, N)

does not fork over N
, so we get a contradiction to the maximality of a
:
Suc
I
() (this explains the role of P in Denition 1.11(5) below).
We are interested in the possible trees N
: I).
Now the tree determines M up to isomorphism, but there are incidental
choices, so two trees may give isomorphic models (for investigating the number
of non isomorphic models it is enough to nd suciently pairwise far trees I).
Here we like to get exact information and in as nitary way as we can. So we
replace (M
, M
, a
) by
_
B
A
_
, where A M
, A + a
B M
, tp(B, M
)
does not fork over A.
Now for I<> we are interested in the possible trees N
: I), over
(N
, N
, a
). But not only dierent trees may be equivalent (giving isomorphic
-prime models) but the other part of the tree, N
: I but ( )) may
apriori cause non equivalent trees to contribute the same toward understanding M.
This is done in [Sh:c, Ch.XII], but here we have to deal with -nite A, B.
The following claim 1.11 really does not add to [Sh:c, Ch.X], it just collects the
relevant information which is proved there, or which follows immediately (paricu-
larly using the parameter (A, B)). We allow here a
/M
- to be not regular, but

this is not serious: we can here deal exclusively with this case and we can omit
this requirement in [Sh:c, Ch.X]; however, this does not eliminate the use of regular
types (in the proof that M is
-prime over every
-decomposition of it).
1.11 Denition. 1) N
, a
: I) is an
-decomposition inside M above (or

for over) the pair
_
B
A
_
(but we may omit the
) if:
(a) I a set of nite sequences of ordinals closed under initial segments
(b) ), 0) I, I) 0) , let I
= I), really a
is meaningless
(c) A N
, B N
0
, N
A
B and dc(a
0
) dc(B),
(d) if = ) I then N
is
-primary over N
, N
is
-prime
over A
(
4
0
1
)

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

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

(e) for I such that k = g() > 1 the type a
/N
(k1)
is orthogonal to
N
(k2)
(f) N
(g) M is
-saturated and N
M for I
(h) if I), then a
: Suc
I
() is (a set of elements realizing over
N
types orthogonal to N
and is) an independent set over N
.
2) We replace inside M by of M if in addition
(i) in clause (h) the set is maximal.
3) N
, a
: I) is an
-decomposition inside M if (a), (d), (e), (f), (g), (h) of

part (1) holds and in clause (h) we allow = ) (call this (h)
+
). We add over A
if A M
<>
.
4) N
, a
: I) is an
-decomposition of M if in addition to 1.11(3) and we

have the stronger version of clause (i) of 1.11(2) by including = ), i.e. we have:
(i)
+
for I, the set a
: Suc
I
() is a maximal subset of M independent
over N
.
We may add over A if A M.
5) If N
, a
: I) is an
-decomposition inside M we let

P(N
, a
: I), M) =
_
p S(M) : p regular and for some I) we have
p is orthogonal to N
but not to N
_
.
As said earlier it is natural to use regular types.
1.12 Denition. 1) We say that N
, a
: I), an
-decomposition inside M,
is J-regular if J I and:
() for each IJ there
2
is c
such that a
ac(N
+ c
)
c
is regular and if ,= ) then

a
+c
a
N
(
)
.
2) We say N
, a
: I) is a regular
-decomposition inside M [of M] if it is

an
-decomposition inside M [of M] which is -regular.

3) We say N
, a
: I) is a regular
-decomposition inside M [of M] over

_
B
A
_
if it is an
-decomposition inside M [of M] over

_
B
A
_
which is )-regular.
2
without loss of generality c
= a
(
4
0
1
)

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

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

12 SAHARON SHELAH
1.13 Claim. 1) Every
-saturated model has an
-decomposition (i.e. of it).

2) If M is
-saturated, N
, a
: I) is an
-decomposition inside M, then

for some J, and N
, a
for JI we have: I J and N
, a
: J) is an
-decomposition of M (even a (JI)-regular one).

3) If M is
-saturated, N
, a
: I) is an
-decomposition of M then M is
-prime and
-minimal
3
over
_
I
N
; if in addition N
, a
: ), 0)) is an
-decomposition inside M above

_
B
A
_
, then N
, a
: I & ( ,= ) 0) ))
is an
- decomposition of M above
_
B
A
_
.
4) If N
, a
: I) is an

_
B
A
_
, then it is an
-decomposition inside M.
5) If N
, a
: I) is an
-decomposition inside M [above

_
B
A
_
], I,
[ I)], = Min : ) / I, =: ), a
MN
,
a
is orthogonal
to M
if
if ,= ), N
M is
-primary over N
+ a
and a
_
_
_
I
N
_
_
(enough to demand a
= and I is independent over a
/N
) then
N
, a
: I ) is an
-decomposition inside M [over

_
B
A
_
].
6) Assume N
, a
: I) is an
-decomposition of M, if p is regular (stationary)

and is not orthogonal to M (e.g. p S(M)) then for one and only one I, there
is a regular (stationary) q S(N
) not orthogonal to p such that: if
is well
dened (i.e. ,= )), then p N
.
7) Assume I =
_
<()
I
, for each we have N
, a
: I
) is an
-decomposition
inside M [above
_
B
A
_
] and for each I for every n < and
) I for
< n, for some we have:
: < n I
(e.g. I
increasing). Then
N
, a
: I) is an
-decomposition inside M [above

_
B
A
_
].
8) In (7), if ,= ) and some
is not -maximal in I and

a
is regular, it is
enough:
1
<
2
< n
<()
[
1
,
2
I
].
9) If N
, a
: I) is an
-decomposition inside M, I
1
, I
2
I are closed under
initial segments and I
0
= I
1
I
2
then
_
_
_
I
1
N
_
_
_
I
0
N
_
_
_
I
2
N
_
_
.
3
here we use NDOP
(
4
0
1
)

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

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

10) Assume that for = 1, 2 that N
, a
: I) is an
-decomposition inside M
,
and for I the function f
is an isomorphism from N
1
onto N
2
and f
. Then
_
I
f
is an elementary mapping; if in addition N
, a
: I) is an
-
decomposition of M
(for = 1, 2) then
_
I
f
can be extended to an isomorphism

from M
1
onto M
2
.
11) If N
, a
: I) is an
- decomposition inside M (above

_
B
A
_
) and M
M
is
-prime over
_
I
N
then N
, a
: I) is an
-decomposition of M
(above
_
B
A
_
).
12) If N
, a
; I) is an
-decomposition inside M/ of M [above

_
M
A
_
] and
a
and N
is
-prime over N
+ a
for I<> [and a
<0>
= a
<0>
or at least dcl(a
<0>
) dcl(B)] then N
, a
: I) is an
-decomposition inside
M/ of M [above
_
B
A
_
].
Proof. 1), 2), 3), 5), 6), 9)-12). Repeat the proofs of [Sh:c, X]. (Note that here
a
/N
is not necessarily regular, a minor change).

4), 7) Check.
8) As Dp(p) > 0 p is trivial, by [Sh:c, ChX,7.2,p.551] and [Sh:c, ChX,7.3].
1.13
We shall prove:
1.14 Claim. 1) If M is
-saturated,
_
B
A
_
(M), then there is N
, a
: I),
an
-decomposition of M above
_
B
A
_
.
2) Moreover if N
, a
: I) satises clauses (a) (h) of Denition 1.11(1), we

can extend it to satisfy clause (i) of 1.11(2), too.
3) If N
, a
: I) is an
_
B
A
_
, M
M is
-prime
over
_
I
N
then:
(a) N
: I) is a
-decomposition of M
(b) we can nd an
-decomposition N
, a
: J) of M such that J I and

[ JI ( ,= ) and 0) )], moreover the last phrase follows from
the previous ones.
4) If in 3)(b) the set JI is countable (nite is enough for our applications), then
necessarily M, M
are isomorphic, even adding all members of an -nite subset

of M
as individual constants.
5) If N
, a
: I) is an
_
B
A
_
, I J and
(
4
0
1
)

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

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

14 SAHARON SHELAH
N
, a
: J) is an
-decomposition of M, M
M is
-prime over
_
I
N
and
_
B
A
_

_
B
1
A
1
_
and B
1
M and c M and
c
B
1
A
1
and
c
B
1
is (stationary and)
regular then
()
c
B
1
{N
:J\I}
N
()
c
B
1
is not orthogonal to some p P(N
, a
: I), M
).
6) If N
, a
: I) is an
_
B
A
_
, then the set P =
P(N
: I), M) depends on
_
B
A
_
and M only (and not on N
: I) or M
when M
M is
-prime over N
: I), recalling:
P = P(N
: I), M) =
_
p S(M) : p regular and for some
I<>, we have :
p is orthogonal to N
but not to N
_
.
So let P(
_
B
A
_
, M) =: P(N
: I), M).
(7) If
B
A
is regular of depth zero or just
b
A

a
B
A
,
b
A
regular of depth zero and M is
-saturated and B M then

(a) for any , we have tp
(
_
B
A
_
, M) depend just on tp
0
(
_
B
A
_
, M)
(b) if
_
B
A
_

_
B
A
_
(M) then tp
(
_
B
A
_
, M) depends just on tp
0
(
_
B
A
_
, M) (and
(A, B, A
, B) but not on M).

8) For < , from tp
(
_
B
A
_
, M) we can compute tp
(
_
B
A
_
, M).
9) If f is an isomorphism from M
1
onto M
2
, A
1
B
1
are -nite subsets of M
1
and f(A
1
) = A
2
, f(B
1
) = B
2
then
tp
(
_
B
1
A
1
_
, M
1
) = tp
(
_
B
2
A
2
_
, M
2
)
(more pedantically tp
(
_
B
2
A
2
_
, M
2
) = f[tp
(
_
B
1
A
1
_
, M
1
)] or considered the A
, B
as
indexed sets).
We delay the proof (parts (1), (2), (3) are proved after 1.22, part (4), (6) after 1.23,
and after it parts (5), (7), (8). Part (9) is obvious.
(
4
0
1
)

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

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

1.15 Denition. 1) If
_
B
A
_
(M), M is
-saturated let P
M
(
B
A
)
be the set P from
Claim 1.14(6) above (by 1.14(6) this is well dened as we shall prove below).
2) Let P
(
B
A
)
=
_
p : p is (stationary regular and) parallel to some p
P
C
eq
(
B
A
)
_
.
1.16 Denition. If N
, a
: J) is a decomposition inside C for = 1, 2 we

say that N
1
, a
: J)
direct
N
2
, a
: J) if:
(a) N
1
N
2
(b) N
2
N
1
: ) J
(c) for J), N
2
is
-prime over N
1
N
2
.
1.17 Claim. 1) M is
-prime over A i M is
-primary over A i M is
-
saturated, A M, M is
-atomic over A (see -1.1(16)) for every I M in-

discernible over A we have: dim(I, M)
0
i M is
-saturated, A M, M is
-atomic over A and for every nite B M and regular (stationary) p S(AB),
we have dim(p, M)
0
.
2) If N
1
, N
2
are
-prime over A, then they are isomorphic over A.

Proof. By [Sh:c, IV,4.18] (see Denition [Sh:c, IV,4.16], noting that we replace F
a
0
by
and that part (4) there disappears when we are speaking on F

a
0
).
1.16
However, we need more specic information saying that minor changes preserve
being
-prime; this is done in 1.18 below, parts of it are essentially done in [Sh
225] but we give full proof.
1.18 Fact. 0) If A is countable, N is
-primary over A then N is
-primary
over .
1) If N is
-prime over , A countable, N

+
is
-prime over N A then N

+
is
-prime over .
2) If N
n
: n < ) is increasing, each N
n
is
-prime over or just
-
constructible over and N
is
-prime over
_
n<
N
n
then N
is
-prime over
, (note that if each N
n
is
-saturated then N
=
_
n<
N
n
).
2A) If N is
-prime over C, a
b N, tp(
b, a) is regular (stationary) and orthogo-

nal to C then dim(tp(
b, a), N)
0
; also if q S(C a) is a non-forking extension
(
4
0
1
)

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

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

16 SAHARON SHELAH
of tp(
b, a) then dim(q, C a) = dim(tp(
b, a), N) =
0
.
2B) If C a
b N and a/
b is a regular type orthogonal to C and q S

g( a)
(N)
is a non-forking extension of a/
b then dim(p (C +

b), N) dim( a/
b, N)
dim(p (C+
b), N) +
0
; moreover, dim(p (C+
b), N) dim( a/
b, N) < dim(p
(C +
b), N)
+
+
0
.
3) If N
2
N
0
N
1
, each N
is
-saturated, N
2
is
-prime over N
0
a, and N
3
is
-prime over N
2
N
1
then N
3
is
-prime over N
1
a.
4) If N
1
N
2
are
-primary over then for some
-saturated N
0
N
1
(neces-
sarily
-primary over ) we have: N

1
, N
2
are isomorphic over N
0
.
5) In part (4), if A N
1
is -nite then we can demand A N
0
.
6) If M
0
is
-saturated, A
M
0
B, M
1
is
-primary over M
0
A then M
1
M
0
B.
7) Assume N
0
N
1
N
2
are
-saturated, N
2
is
-primary over N
1
+ a and
a
N
1
N
0
(and a / N
1
). If N
0
N
0
, N
0
N
1
N
1
, N
0
N
0
and N
1
is
-
primary over N
0
N
1
, A
1
N
1
, A
2
N
2
are -nite and tp
(A
2
, N
1
) does not fork
over A
1
, then we can nd a
, N
2
such that: N
2
is
-saturated,
-primary over
N
1
+a
, N
1
N
2
N
2
, N
1
1
N
2
and N
2
is
-primary over N
1
N
2
and A
2
N
2
.
8) Assume N
0
N
0
N
1
and a N
1
and N
1
is
-prime over N
0
+a and
a
N
0
N
0
and A
0
N
0
, A
1
N
1
are -nite and tp
(A
1
, N
0
) does not fork over A
0
then we
can nd a
, N
1
such that a
, N
0
N
1
N
1
, N
0
N
0
, N
1
is
-prime over
N
0
+ a and N
1
is
-prime over N
0
+ N
1
and A
1
N
1
.
9) If N
1
is
-prime over and A B N

1
and A, B are -nite, then we can
nd N
0
such that: A N
0
N
1
, N
0
is
-prime over , A N
0
, B
A
N
0
, and N
1
is
-prime over N
0
B.
10) If N
0
is
-prime over A and B N

0
is -nite, then N
0
is
-prime over AB
(and also over A
if A A
acl(A)).
1.19 Remark. In the proof of 1.18(1)-(6),(10) we do not use T has NDOP.
Proof. 0) There is a
: <
, a list of members of N in which every member

of NA appears such that for < () we have: tp(a
, A a
: < ) is
-isolated (which means just F

a
0
-isolated).
[Why? by the denition of N is
-primary over A). Let b

n
: n < list A
(if A = the conclusion is trivial so without loss of generality A ,= , hence we
can nd such a sequence b
n
: n < )). Now dene
= + and b
+
= a
(
4
0
1
)

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

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

for <
. So b
: <
lists the elements of N (possibly with repetitions,

remember A N and check). We claim that tp(b
, b
: < ) is F
a
0
-isolated
for <
.
(Why? if , let
= (so <
), now the statement above means

tp(a
, A a
: <
) is F
a
0
-isolated which we know; if < this statement
is trivial)]. By the denition of F
a
0
-primary, clearly b
: < + ) exemplify
that N is F
a
0
- primary over .
1) Note
()
1
if N is
-primary over and A N is nite then N is
-primary over A
[why? see [Sh:c, IV,3.12](3),p.180 (of course, using [Sh:c, IV, Table 1,p.169]
for F
a
0
]
()
2
if N is
-primary over , A N is nite and p S

m
(N) does not fork
over A and p A is stationary then for some a
: < we have: a
N
realize p, a
: < is independent over A and p (A

_
<
a
) p
[why? [Sh:c, IV,proof of 4.18] (i.e. by it and [Sh:c, 4.9](3),4.11) or let N
be
-primary over A
_
<
a
and note: N
is
-primary over A (proof

like the one of 1.18(0)) but also N is
-primary over A so by uniqueness

of the
-primary model N
is isomorphic to N over A, so without loss of

generality N
= N; and easily N
is as required].
Now we can prove 1.18(1), for any c
>
A, we can nd a nite B
1
a
N
such that tp( c, N) does not fork over B
1
c
, let

b
c

>
N realize stp( a, B
1
a
) and let
B
c
= B
1
c

b
c
, so tp( c, N) does not fork over B
c
and tp( c, B
c
) is stationary, hence
we can nd a
c
: < as in ()
2
(for tp( c, B
c
)). Let
A
= B
c
: c
>
A a
c
: c
>
A and < , so A
is a countable subset
of N and tp
(A, A
) tp(A, N) = stp(A, N). As N is
-primary over we can

nd a sequence d
: <
) and w
: <
) such that N = d
: <
and w
is nite and stp(d
, d
: w
) stp(d
, d
: < ) and
< d
,= d
.
We can nd a countable set W
such that A
: W and
W w
W. Let A
= a
: W. By [Sh:c, IV,2,3] without loss of

generality W is an initial segment of
.
Easily
<
& / W stp(d
, d
: w
) stp(d
, A d
: < ).
(
4
0
1
)

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

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

18 SAHARON SHELAH
As N
+
is
-primary over NA we can nd a list d
: [
) of N
+
(NA)
such that tp(d
, N A d
: [
)) is
-isolated. So
d
: / W, <
) exemplies that N
+
is
-primary over A A
, hence by
1.18(0) we know that N
+
is
-primary over .
2) We shall use the characterization of N is F
a
0
-prime over A in 1.17, more
exactly we use the last condition in 1.17(1) for A = , M = N
. Clearly N
is
-
saturated (as it is
-prime over
_
n<
N
n
). Suppose B N
is nite and p S(B)

is (stationary and) regular.
Case 1: p not orthogonal to
_
n<
N
n
.
So for some n < , p is not orthogonal to N
n
, hence there is a regular p
1
S(N
n
)
such that p, p
1
are not orthogonal. Let A
1
N
n
be nite such that p
1
does not
fork over A and p
1
A
1
is stationary. So by [Sh:c, V,2] we know dim(p, N
) =
dim(p
1
A
1
, N
), hence it suces to prove that the latter is

0
. Now this holds
by [Sh:c, V,1.16](3),p.237 or immitate the proof of ()
2
above.
Case 2: p is orthogonal to
_
n<
N
n
.
Note that if each N
n
is
-prime then
_
n<
N
n
is
-saturated hence N =
_
n<
N
n
hence this case does not arise. Let A =
_
n<
N
n
, so dim(p, N)
0
follows from
2A) below.
Alternatively (and work even if we replace N
by a set A
n
, F
a
0
-constructible over
, see below).
2A) By 2B).
2B) The rst inequality as immediate (as T is superstable and a,
b are nite), so
let us concentrate on the second. Let B C be a nite set such that tp
( a
b, C)
does not fork over B and stp
( a
b, B) stp
( a
b, C). Recall q S(N) extend

a/
b and do not fork over

b, let b
C realize q and let q

1
= stp(
, B
b) and
q
2
= stp(
, C
b). Now by the assumption of our case q

1
is orthogonal to tp
(C, B)
hence (see [Sh:c, V,3]) q
1
q
2
and let a
: <
(q
1
(
b B))(N) be
a maximal set independent over C +

b, so [
[ dim( a/(C +

b), N) and q
(C
b a
: <
) q. Also clearly stp
(a
: <
b B) stp
(a
:
<
b C). Together dim(q

1
, N) [
[ and as [B[ <

0
=
r
(T) clearly
dim( a/
b, N) <
0
+ dim(q
1
, N)
+
, so we are done.
We can use a dierent proof for part (2), note:
1
if = cf()
r
(T) and B
is F
a
-constructible over A for < ,

(
4
0
1
)

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

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

and < < B
then
_
<
B
is F
a
-constructible over A
[why? see [Sh:c, IV,3], [Sh:c, IV,5.6,p.207] for such arguments, assume
A
= A, a
i
: i < i
), B
i
: i < i
)) is an F
a
-construction of B
over A.
Without loss of generality i < j < i
i
,= a
i
, and choose by induction
on , u
: < ) such that: u
, u
increasing continuous in i, u
0
=
, [u
+1
u
[ , u
is A
-closed and < < implies a
j
: j u
j
: j u
and tp
(a
i
: i u
, A a
i
: i < i
) does not fork

over A a
i
: i u
. Now nd a list a
j
: j < j
) such that for each

, j : a
j
a
i
: i u
for some < , < is an initial segment
of j
and
+1

+ .]
We use
1
for =
0
.
So each N
n
is
-constructible over hence

_
n<
N
n
is
-constructible over and

also N
is
-constructible over
_
n<
N
n
hence N
is
-constructible over . But

N
is
-saturated hence N
is
-primary over . Alternatively use: if B is F

a
-
constructible over A,
r
(T) and I is indiscernible over A, [I[ > then for some
J I of cardinality , IJ is an indiscernible set over B.
3) Suppose N
3
is
-saturated and N
1
+ a N
3
. As N
2
is
-prime over N
0
+ a
and N
0
+ a N
1
+ a N
3
we can nd an elementary embedding f
0
of N
2
into N
3
extending id
N
0
+ a
. By [Sh:c, V,3.3], the function f
1
= f
0
id
N
1
is an elementary
mapping and clearly Dom(f
1
) = N
1
N
2
. As N
3
is
-prime over N
1
N
2
and f
1
is an elementary mapping from N
1
N
2
into N
3
which is an
-saturated model
there is an elementary embedding f
3
of N
3
into N
3
extending f
2
. So as for any
such N
3
there is such f
3
, clearly N
3
is
-prime over N
1
+ a, as required.
4) Let N
0
be
0
-prime over and let p
i
: i < S(N
0
) be a maximal family of
pairwise orthogonal regular types. Let I
i
= a
i
n
: n < C be a set of elements
realizing p
i
independent over N
0
and let I =
_
i<
I
i
and N
1
be F
a
0
-prime over N
0
I.
Now
() if a,
b N
1
and a/
b is regular (hence stationary), then dim( a/
b, N
1
)
0
.
[Why? If a/
b N
0
then dim( a/
b, N
1
)
0
by part (2A) and the choice of the p
i
and I
i
for i < . If a/
bN
0
, then for some

b
N
0
realizing stp(
b a, ), we have
a
a/b hence dim( a/b, N
1
) = dim( a
, N
1
), so without loss of generality

b a
N
0
, similarly without loss of generality there is i() < such that a/
b p
i()
and
p
i()
does not fork over
b now easily dim( a/
b, N
1
) = dim( a/
b, N
0
)+ dim(p
i()
, N
0
)
0
+
0
=
0
(see [Sh:c, V,1.6](3)). So we have proved ()].
(
4
0
1
)

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

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

20 SAHARON SHELAH
Now use 1.17(1) to deduce: N
1
is F
a
-prime over hence (by uniqueness of
-prime
model, 1.17(2)) N
= N
1
.
By renaming without a loss of generality N
1
= N
1
. Now
()() (N
1
, c)
cN
0
, (N
2
, c)
cN
0
are
-saturated and
() if a C,
b N
, a/
b a regular type and a
b
(N
0
+

b) (for = 1 or = 2),
then
dim( a/(
b N
0
), N
) =
0
.
[Why? Remember that we work in (C
eq
, c)
cN
0
. The
-saturated follows from

the second statement.
Note: dim( a/(
b N
0
), N
) dim( a/
b, N
)
0
(rst inequality by monotonicity,
second inequality by 1.17(1) and the assumption N
is
-prime over ). If a/
b
is not orthogonal to N
0
then for some i < we have p
i
( a/
b) so easily (using
N
is
-saturated) we have dim( a/(
b N
0
), N
) = dim(p
i
, N
) |I
i
| =
0
;
so together with the previous sentence we get equality. Lastly, if a/
b N
0
by
part (2B) of 1.18, we have dim( a/(
b N
0
), N
) <
0
dim( a/
b, N
) <
0
which
contradicts the assumption N
is
-saturated.] So we have proved () hence by

1.17(1) we get N
1
, N
2
0
as required.
5) This is proved similarly as if N is
-prime over A and B N is -nite then

N is
-prime over A + B and also over A
if A + B A
acl(A + B), see part

(10).
6) By [Sh:c, V,3.2].
7) First assume that A
2
N
1
and a/N
1
is regular. As N
1
is
-prime over N
0
N
1
and as T has NDOP (i.e. does not have DOP) we know (by [Sh:c, X,2.1,2.2,p.512])
that N
1
is
-minimal over N
0
N
1
and
a
N
1
0
or to N
1
.
But a/N
1
N
0
by an assumption, so a/N
1
1
hence there
is a regular p
S(N
1
) not orthogonal to
a
N
1
hence (by [Sh:c, V,1.12,p.236]) p
is realized say by a
N
2
. By [Sh:c, V,3.3], we know that N
2
is
-prime over
N
1
+a
. We can nd N
2
which is
-prime over N
1
+a
and N
2
which is
-prime
over N
1
N
2
hence by part (3) of 1.18 we know that N
2
is
-prime over N
1
+a
so
by uniqueness, i.e. 1.17(1), without loss of generality N
2
= N
2
hence we are done.
In general by induction on choose N
2,
such that N
2,0
is
-prime over N
1

A
2
, N
2,
is increasing with and N
1
1
N
2,
. Easily for some , N
2,
is dened
but not N
2,+1
. Necessarily N
2
is
-prime over N
1
N
2,
. Lastly let a
2,
be such that tp(a, N
1
N
2,
) does not fork over N
1
+ a
. Easily N
2,
is
-prime
over N
1
+ a
by (1.17(1)).
8) Similar easier proof.
9) Let N
0
be
-prime over A such that B
A
N
0
, and let N
1
be
-prime over
(
4
0
1
)

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

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

N
0
B. By 1.18(1), we know that N
1
is
-prime over , and by 1.18(10) below N
1
is
-prime over AB, hence by 1.17(2) we know that N
1
, N
1
are isomorphic over
A B hence without loss of generality N
1
= N
1
and so N
0
= N
0
is as required.
10) By [Sh:c, IV,3.12](3),p.180.
1.18
1.20 Fact. Assume N
1
, a
: I)
direct
N
2
, a
: I) (see Denition 1.16)

and A B N
1
<0>
and
I
N
2
M.
(1) If = ) I, then N
2
N
1
N
1
and even N
2
N
1
_
_
_
_
I
N
1
_
_
_
; and
I implies N
2
N
1
_
_
_
_
T
N
1
_
_
_
.
(2) N
2
, a
: I) is an

_
B
A
_
i
N
1
, a
: I) is an

_
B
A
_
.
(3) Similarly replacing

_
B
A
_
by
_
B
A
_
.
Proof. 1) We prove the rst statement by induction on g(). If =<> this is
clause (b) by the Denition 1.16 and clause (d) of Denition 1.11(1) (and [Sh:c,
V,3.2]). If ,=<>, then
a
N
1
(
)
(by condition (e) of Denition 1.11(1)).
By the induction hypothesis N
2
(
N
1
(
)
N
1
and we know N
2
is
-primary over
N
2
(
)
N
1
; we know this implies that no p S(N

1
) orthogonal to N
1
is realized
in N
2
hence
a
N
1
N
2
N
1
, so
a
N
1
N
2
hence
N
1
N
1
N
2
N
1
hence N
1
N
1
N
2
as required.
The other statements hold by the non-forking calculus (remember if = ) I
then use tp(N
1
: I, N
1
) is orthogonal to N
1
or see details in the proof

of 1.21(1)()).
2) By Denition 1.16, for = 1, 2 we have: N
, a
: I) is a decomposition
inside C and by assumption
I
N
1
N
2
M. So for = 1, 2 we have to prove

N
, a
: I) is an
-decomposition inside M for

_
B
A
_
assuming this holds for
1 . We have to check Denition 1.11(1).
(
4
0
1
)

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

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

22 SAHARON SHELAH
Clauses 1.5(1)(a),(b) for holds because they hold for 1 .
Clause 1.5(1)(c) holds as by the assumptions A B N
1
<0>
N
2
<0>
, A N
1
<>
and N
1
<0>
N
1
<>
N
2
<>
.
Clauses 1.5(1)(d),(e),(f),(h) holds as N
, a
: I) is a decomposition inside C
(for = 1 given, for = 2 easily checked).
Clause 1.5(1)(g) holds as
N
1
N
2
M is given and M is
-saturated.
3) First we do the only if direction; i.e. prove the maximality of N
1
, a
: I)
as an

_
B
A
_
(i.e. condition (i) from 1.11(2)), assuming
it holds for N
2
, a
: I). If this fails, then for some I<> and

a M, a
<>
: < > I a is independent over N
1
and
a / a
: ) I and
a
N
1
N
1
. Hence, if
) I for < k then

a = a)a
<
>
: < k) realizes over N
1
a type orthogonal to N
1
, but
N
1
N
1
, N
1
N
2
and N
1
N
1
N
2
(see 1.20(1), hence (by [Sh:c, V,2.8])

tp( a, N
2
) N
2
hence aa
: < k is independent over N

2
but k,
)
I for < k were arbitrary so a a
: ) I is independent over N
2
contradicting condition (i) from Denition 1.11(2) for N

2
, a
: I).
For the other direction use: if the conclusion fails, then for some I<>
and
a MN
2
a
<>
: < > I the set a
: ) I a is inde-
pendent of N
2
and tp(a, N
2
2
; let N
M be
-prime over
N
2
+ a. But N
2
is
-prime over N
1
N
2
(by the denition of -direct) so by

NDOP tp(a, N
2
) N
1
hence there is a regular q S(N

1
) such that q tp(a, N

2
).
Hence some a
realizes q, clearly a
<>
: < > Ia
is independent
over N
2
(and a
/ a
1
<>
: < > I) hence over (N
2
, N
1
) and easily we get

contradiction.
1.20
1.21 Fact. Assume N
1
, a
1
: I) is an
-decomposition inside M.
1) If N
1
<>
N
2
<>
M, N
2
is
-prime over and

N
2
<>
N
1
<>
a
1
<>
:< > I then
()
_
_
N
2
<>
N
1
<>
_
I
N
1
_
_
and
(
4
0
1
)

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

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

() we can nd N
2
( I<>) such that N

2
M, and
N
1
, a
1
: I)
direct
N
2
, a
1
: I).
2) If Cb
a
1
<>
N
1
<>
N
0
<>
N
1
<>
or at least N
0
<>
N
1
<>
and
a
1
<>
N
1
<>
N
0
<>
whenever
< > I then we can nd N
0
M and a
0
(for I<>) such that

N
0
, a
0
: I)
direct
N
1
, a
0
: I).
3) In part (2), if in addition we are given B
: I) such that B
is an -nite
subset of N
, tp
(B
, N
) does not fork over B
and B
<>
N
0
<>
then we can
demand in the conclusion that I B
N
0
.
Proof. 1) For proving () let
i
: i and without loss of generality for some
we have
i
< >:< > I i [1,
). Now we prove by induction on [1, i
)
that N
2
<>
N
1
<>
N
1
i
: i < . For = 1 this is assumed. For limit use the local
character of non-forking.
If = + 1 [1,
), then by repeated use of [Sh:c, V,3.2] (as a
0
: j [1, )
is independent over (N
1
<>
, N
2
<>
) and N
1
<>
is
-saturated and N
1
j
(j [1, )) is
-prime over N
1
<>
+a
j
) we know that tp(a
, N
2
<>
_
i<
N
1
i
) does not fork over
N
1
<>
. Again by [Sh:c, V,3.2], the type tp
(N
1
, N
2
<>

_
i<
N
1
i
N
1
<>
hence
_
i<
N
1
N
1
<>
N
2
<>
and use symmetry.
Lastly, if + 1 [
, i
), tp(a
, N
1
<>
and even to
N
1
(
so again by non-forking and [Sh:c, V,3.2] we can do it, so clause () holds.

For clause (), we choose N
2
i
for i [1, i
) by induction on i < i
such that
N
2
i
M is
-prime over N
2
i
N
1
i
. By the non-forking calculus we can check
Denition 1.7.
2) We let
i
: i < i
be as above, now we choose N

0
i
, a
0
i
by induction on i [1, i
)
such that:
() N
0
i
N
1
i
and N
1
N
0
i
N
0
i
and N
1
i
is
-prime over N
0
i
N
1
i
() a
0
i
N
0
i
and N
0
i
is
-prime over N
0
i
+ a
0
i
.
The induction step has already been done: if g(
i
) > 1 by 1.18(7) and if g(
i
) = 1
(
4
0
1
)

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

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

24 SAHARON SHELAH
by 1.18(8).
3) Similar.
1.21
1.22 Fact. 1) If N
1
, a
: I)
direct
N
2
, a
: I) and both are
-
decompositions of M above
_
B
A
_
, then
P(N
1
, a
1
: I), M) = P(N
2
, a
2
: I), M).
Proof. By Dention 1.11(5) it suces to prove, for each I<> that
() for regular p S(M) we have
p N
1
& p N
1
p N
2
& p N
2
.
Now consider any regular p S(M): rst assume p N
1
& p N
1
where
I<> so p N
2
(as N
1
N
2
and p N
1
) and we can nd a regular

q S(N
1
) such that q p; so as p N
1
also q N
1
, now q N
2
(as
N
1
N
1
N
2
and q N
1
see [Sh:c, V,2.8]) hence p N

2
.
Second, assume p N
2
& pN
2
where I<>; remember N

1
, N
1
, N
2
, N
3
are
-saturated, N
1
N
1
N
2
and N
2
is
-prime over N
1
N
2
and T does not have

DOP. Hence N
2
is
-minimal over N
1
N
2
and every regular q S(N

2
) is not
orthogonal to N
1
or to N
2
. Also as p N
2
there is a regular q S(N

2
) not or-
thogonal to p, so as p N
2
also q N
2
; hence by the previous sentence q N

1
hence p N
1
. Lastly, as p N
2
and N
1
N
2
clearly p N
1
, as required.
1.22
At last we start proving 1.14.
Proof of 1.14. 1) Let N
0
C be
-primary over A, without loss of generality

N
0
A
B (but not necessarily N
0
M), and let N
1
be
-primary over N
0
B.
Now by 1.18(0) the model N
0
is
-primary over and by 1.18(1) the model N

1
is
-primary over hence (by 1.18(10)) is
-primary over B, hence without loss of

generality N
1
M. Let N
<>
=: N
0
, N
<0>
= N
1
, I = <>, < 0 > and a
<0>
= B.
More exactly a
is such that dcl(a
) = dcl(B). Clearly N
, a
: I) is an

_
B
A
_
. Now apply part (2) of 1.14 proved below.
(
4
0
1
)

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

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

2) By 1.13(4) we know N
, a
: I) is an
-decomposition inside M, by 1.18(2)

we then nd J I and N
, a
for JI such that N
, a
: I) is an
-
decomposition of M. By 1.18(3), N
, a
: J
) is an
-decomposition of M
above
_
B
A
_
where J
=: J : =<> or 0) J.
3) Part (a) holds by 1.13(2),(3). As for part (b) by 1.13(2) there is N
, a
: J),
an
-decomposition of M with I J; easily [0) J I].
1.14(1),(2),(3)
1.23 Fact. If N
, a
: I
) are
-decompositions of M above
_
B
A
_
, for = 1, 2
and N
1
<>
= N
2
<>
then P(N
1
, a
1
: I
1
), M) = P(N
2
, a
2
: I
2
), M).
Proof. By 1.14(3)(b) we can nd J
1
I
1
and N
1
, a
1
for J
1
I
1
such that
N
1
, a
1
: J
1
) is an
-decomposition of M and moreover we have J

1
I
1
,= ) & (0) ). Let J

2
= I
2
(J
1
I
1
) and for J
2
I
2
let a
2
=: a
1
,
N
2
=: N
1
. Easily N
2
, a
2
: J
2
) is an
-decomposition of M. By 1.13(6) we
know that for every regular p S(M) there is (for = 1, 2) a unique (p, ) J
such that p N
(p,)
& p N
(p,)
(note )
meaningless). By the uniqueness

of (p, ), if (p, 1) J
1
I
1
then as it can serve as (p, 2) clearly it is (p, 2) so
(p, 2) = (p, 1) J
1
I
1
= J
2
I
2
; similarly (p, 2) J
2
I
2
(p, 1) J
1
I
1
and
(p, 1) = ) (p, 2) = ). So
() (p, 1) I
1
) (p, 2) I
2
).
But
() (p, ) I
) p P(N
, a
: I
), M).
Together we nish.
1.23
We continue proving 1.14.
Proof of 1.14(4). Let A
be -nite, so we can nd an -nite B
:
I such that stp(A
, B
) stp(A
, N
: I). Hence, there is a nite

non empty I
I such thatd <> I
, I
is closed under initial segments and B
: I
, so of course stp
(A
, N
: I
) stp(A
, N
: I).
We can also nd B
: I
) such that B
is an -nite subset of N
, B
=
acl(B
) and B
: I
, ,=<> a
and if I
then B
and tp
(B
, N
) does not fork over B
. Without loss of generality B B
<0>
.
For II
let B
= B
where < g() is maximal such that I
, such
exists as g() is nite and <> I
.
(
4
0
1
)

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

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

26 SAHARON SHELAH
Let N
1
= N
and a
1
= a
for I and without loss of generality J ,= I hence

JI ,= .
Let N
2
<>
M be
-prime over
_
J\I
N
; letting JI =
i
: i ,i
(for i i
) increasing continuous,
N
2
<>,0
= N
<>
and N
2
<>,i+1
is
-prime over N
2
<>,i
N
i
hence over N
2
<>,i
+ a
i
.
Lastly, without loss of generality N
2
<>,i
= N
2
<>
.
By 1.18(1),(2) we know N
2
<>
is
-primary over and (using repeatedly 1.18(6)

+ nite character of forking) we have N
2
<>
N
1
<>
a
<0>
. By 1.18(4)
(with N
1
<>
, N
2
<>
, B
<>
Cb(a
<>
/N
1
<>
) here standing for N
1
, N
2
, A there) we
can nd a model N
0
<>
such that a
<0>
N
0
<>
N
1
<>
and Cb(a
<>
/N
1
<>
) B
<>

N
0
<>
, N
0
<>
N
1
<>
, N
0
<>
is
-primary over and N

1
<>
, N
2
<>
are isomorphic over
N
0
<>
. By 1.21(1) we can for I choose N
2
M with N
1
N
2
and N
1
, a
1
:
I)
direct
N
2
, a
0
: I). Similarly, by 1.21(2) (here Suc

I
(<>) = 0))
we can choose an
-decomposition N
0
, a
0
: I) with N
0
, a
0
: I)
direct
N
1
, a
1
: I). Moreover, we can demand I
N
0
using 1.21(3).
By 1.13(12) + 1.14(3) we know that N
1
, a
0
: I) is an
-decomposition
of M
and easily N
2
, a
0
: I) is an
-decomposition of M. Now choose

by induction on I an isomorphism f
from N
1
onto N
2
over N
0
such that
f
and I
= id
B
. For =<> we have chosen

N
0
such that N
1
, N
2

0
. For the induction step note that

f
(
)
id
N
0
is an elementary mapping as N
2
(
N
0
(
)
N
0
and f
(
)
id
N
0
can
be extended to an isomorphism f
from N
1
onto N
2
as N
is
-primary (in fact

even
-minimal) over N
)
N
0
for = 1, 2 (which holds easily). If I
there is no problem to add f
= id
B
. Now by 1.13(3) the model M
is
-saturated and
-primary and
-minimal over
_
J
N
=
_
I
N
1
; similarly M is
-primary over
_
I
N
2
. Now
_
is an elementary mapping from

_
I
N
1
onto
_
I
N
2
hence can be extended to an isomorphism f from M
into M. Moreover
as stp
(A
, B
: I
) stp(A
, N
1
: I), by [Sh:c, CH.XII,4] we have

tp
(A
, B
: I
tp(A
, N
1
: I hence tp
(A
, B
: I
) has
a unique extension as a complete type over N
1
: I hence over N
2
: I
(
4
0
1
)

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

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

so without loss of generality f A
= id
A
. By the
-minimality of M over
_
I
N
(see 1.13(3)), f is onto M, so f is as required.

1.14(4)
We delay the proof of 1.14(5).
Proof of 1.14(6). Let N
, a
: I
) for = 1, 2, be
-decompositions of M
above
_
B
A
_
, so dcl(a
<>
) = dcl(B). Let p S(M), and assume that p P(N
1
, a
1
:
I
1
), M), i.e. for some I
1
<>, (p
) and p
. We shall prove
that the situation is similar for = 2; i.e. p P(N
2
, a
2
: I
2
), M); by
symmetry this suces.
Let n = g(), choose B
: n) and d such that:

() A B
0
,
() B B
1
,
() a
N
1
, for n
() B
+1
N
1
,
()
B
+1
B
+a
1
(+1)
B
+1
N
1
+a
1
(+1)
,
() d B
n
,
d
B
n
\{d}
is regular p, (hence B
n1
)
() B
is -nite.
[Why such B
: n) exists? We prove by induction on n that for any I of

length n and -nite B
, there is B
: n) satisfying () (), () such

that B
B
n
. Now there is p
S(N
1
) regular not orthogonal to p, let B

1
N
1
be an -nite set extending Cb(p
). Applying the previous sentence to , B

1
we get
B
: n), let d N
realize p
B
n
.
Now as n > 0, tp(d, B
n
) N
hence tp(d, B
n
) B
n1
, hence
tp(d, B
n
) tp
(N
, B
n
), hence as tp(d, B
n
) is stationary, by [Sh:c, V,1.2](2),p.231,
the types tp(d, B
n
),tp
(N
, B
n
) are weakly orthogonal so
tp(d, B
n
) tp(d, N
B
n
) hence
B
n
+d
B
n1
+a
1
B
n
+d
N
1
+a
1
.
Now replace B
n
by B
n
d and we nish].
Note that necessarily
()
+
B
n
B
m
N
1
m
for m n.
[Why? By the non-forking calculus].
()
+ B
n
B
m
+a
1
(m+1)
a
B
m
for m < n.
[Why? As N
1
m
is
-saturated].
(
4
0
1
)

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

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

28 SAHARON SHELAH
Choose D
N
2
<>
nite such that
B
n
N
2
<>
+B
does not fork over D
+ B.
[Note: we really mean D
N
2
<>
, not D
N
1
<>
].
We can nd N
3
<>
,
-prime over such that A N

3
<>
N
2
<>
and D
A
N
3
<>
and N
2
<>
is
-prime over N
3
<>
D
(by 1.18(9)). Hence B

n
A
N
3
<>
and B
n
B
N
3
<>
(by the non-forking calculus). As tp
(B, N
2
<>
) does not fork over A N
3
<>
N
2
<>
by 1.21(2) we can nd N
3
, a
3
(for I
2
<>), such that N
3
, a
3
: I) is an

_
B
A
_
and N
3
, a
3
: I
2
)
direct
N
2
, a
3
: I
2
)
and a
3
<0>
= a
2
<0>
(remember dcl(a
2
<0>
) = dcl(B)). By 1.20(2) we know N
3
, a
3
:
I
2
) is an
_
B
A
_
.
By 1.22 it is enough to show p P(N
3
, a
3
: I
2
), M). Let N
4
<>
N
2
<>
M
be
-prime over N
3
<>
B
0
. Now by the non-forking calculus B
A
(N
3
<>
B
0
) [why?
because
(a) as said above B
n
B
N
3
<>
but B
0
B
n
so B
0
B
N
3
<>
, and
(b) as B
A
N
1
<>
and B
0
N
1
<>
we have B
A
B
0
so B
0
A
B
hence (by (a) + (b) as A B)
(c)
B
0
N
3
<>
+B
does not fork over A,
also
(d) B
A
N
3
<>
(as A N
3
<>
N
2
<>
and tp(B, N
2
<>
) does not fork over A)
putting (c) and (d) together we get
(e)
A
B
0
, B, N
3
<>
hence the conclusion].

Hence B
N
3
<>
B
0
so B
N
3
<>
N
4
<>
(by 1.18(6)) and so (as N
3
<0>
is
-prime over
N
3
<>
+dcl(a
3
<>
) = N
3
<>
+dcl(B)) we have N
4
<>
N
3
<>
N
3
<0>
and by 1.21(1) we can
choose N
4
M (for I
2
<>), such that N
4
, a
3
: I
2
)
direct
N
3
, a
3
:
I
2
). So by 1.20(1) N
4
, a
3
: I
2
) is an
_
B
A
_
hence
a
3
<0>
/N
4
does not fork over A but A B

0
N
4
<>
so a
3
<>
/N
4
does not fork over B

0
(
4
0
1
)

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

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

and by 1.22 it is enough to prove p P
_
N
4
, a
3
: I
2
), M
_
. Now as said above
B
N
3
<>
N
4
<>
and B
A
N
3
<>
so together B
A
N
4
<>
, also we have A B
0
N
4
<>
,
hence B
B
0
N
4
<>
and
B
n
B
0
+B

B
n
B
0
+a
3
<0>
a
B
0
(by ()
+
above) but a
3
<>
B
0
N
4
<>
hence
B
n
N
4
<>
+a
3
<0>
is
-isolated. Also letting B
n
= B
n
d we have
B
n
N
4
<>
+a
3
<>
is
-isolated and
d
B
B
0
(by clause ()), and clearly d
n
(N
4
<>
B
n
) so
d
B
n
N
4
<>
. Hence we can nd
N
5
, a
5
: I
5
) an
_
B
A
_
such that N
5
<>
= N
4
<>
,
dcl(B) = dcl(a
5
<0>
), B
n
d N
5
<0>
and d = a
5
<0,0>
(on d see clause () above)
so d
B
n
N
5
<0>
.
By 1.23 it is enough to show p P(N
5
, a
5
: I
5
), M) which holds trivially
as tp(d, B
n
d) witness.
1.14(6)
Proof of 1.14(5). By 1.8, with A, B, A
1
, B
1
here standing for A
1
, B
1
, A
2
, B
2
there,
we know that there are B
: n), c
: 1 < n) as there. By 1.18(9) we can

choose N
1
<>
such that B
0
N
1
<>
, N
1
<1>
B
0
B
n
, N
1
<>
is
-primary over . Then

we choose N
1
, a
1
:
_
<>, < 0 >, < 0, 0 >, . . . , < 0, . . . , 0 >
. .
n
_
), (where
N
1
< 0, . . . , 0 >
. .
n
M and > 0 a
1
< 0, . . . , 0 >
. .
= c
, B
g()
N
1
and we choose
N
1
by induction on g() being
-prime over N
1
a
1
hence a
1
/N
1
does not
fork over B
g(
)
hence N
1
is
-prime also over N

1
+ B
g()
. So N
1
, a
1
:
<>, . . . ) is an

_
B
1
A
1
_
. Now apply rst 1.14(2) and
then 1.14(6).
Proof of 1.14(7). Should be easy. Note that
()
1
for no
_
B
_
do we have
_
B
A
_

b
_
B
_
Why? By the denition of Depth zero.
()
2
if
_
B
A
_
<
a
_
B
_
, then also
_
B
_
satises the assumption.
Hence
(
4
0
1
)

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

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

30 SAHARON SHELAH
() for no
_
B
1
A
1
_
,
_
B
2
A
2
_
do we have
_
B
A
_

a
_
B
1
A
1
_
<
b
_
B
2
A
2
_
.
[Why? As also
_
B
1
A
1
_
satises the assumption].
Now we can prove the statement by induction on for all pairs
_
B
A
_
satisfying the
assumption. For = 0 the statement is a tautology. For limit ordinal reread
clause (c) of Denition 1.9(1). For = +1, reread clause (b) of Denition 1.9(1):
on tp
(
_
B
A
_
, M) use the induction hypothesis also for computing Y
1,B
A,B,M
(and reread
the denition of tp
0
, in Denition 1.9(1), clause (a)). Lastly Y
2,
A,B,M
is empty by
() above.
Proof of 1.14(8),(9). Read Denition 1.9.
1.14(5),(7),(8),(9)
Discussion. In particular, the following Claim 1.26 implies that if N
, a
: I)
is an
_
B
A
_
and M
is
-prime over N
: I
then
_
B
A
_
has the same tp
in M and M
.
1.24 Claim. 1) Assume that M
1
M
2
are
-saturated,
_
B
A
_
(M
1
). Then the
following are equivalent:
(a) if p P
__
B
A
_
, M
1
_
(see 1.14(6) for denition; so p S(M
1
) is regular) then p is not realized
in M
2
(b) there is an
-decomposition of M
1
above
_
B
A
_
, which is also an
-decomposition
of M
2
above
_
B
A
_
(c) every
-decomposition of M
1
above
_
B
A
_
is also an
-decomposition of M
2
above
_
B
A
_
.
2) If M is
-saturated,
_
B
1
A
1
_

_
B
2
A
2
_
are both in (M) then P
__
B
2
A
2
_
, M
_

P
__
B
1
A
1
_
, M
_
.
3) The conditions in 1.24(1) above implies
(d) p P
__
B
A
_
, M
2
_
p M
1
.
(
4
0
1
)

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

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

Proof. 1) (c) (b).
By 1.14(1) there is an
-decomposition of M
1
above
_
B
A
_
. By clause (c) it is
also an
-decomposition of M
2
above
_
B
A
_
, just as needed for clause (b).
(b) (a)
Let N
, a
: I) be as said in clause (b). By 1.14(3)(b) we can nd J

1
, I J
1
and N
, a
(for J
1
I) such that N
, a
: J
1
) is an
-decomposition of M
1
and J
1
I (0) > 0. Then we can nd J
2
, J
1
J
2
and N
, a
(for J
2
J
1
)
such that N
: J
2
) is an
-decomposition of M
2
(by 1.14(2)). By 1.14(3)(b),
J
2
I (0) > 0. So I) Suc
J
2
() = Suc
I
(), hence
() if I) and q S(N
) is regular orthogonal to N
then the station-

arization of q in S(M
1
) is not realized in M
2
.
Now if p P
__
B
A
_
, M
1
_
then p S(M
1
) is regular and (see 1.14(1), 1.11(5)) for
some I), p N
, pN
, so there is a regular q S(N
) not orthogonal to
p. Now no c M
2
realizes the stationarization of q over M
1
(by () above), hence
this applies to p, too.
(a) (c)
Let N
, a
: I) be an
-decomposition of M
1
above
_
B
A
_
. We can nd
N
, a
: J) an
-decomposition of M
1
such that I J and JI
(0) > 0 (by 1.14(3)(b)), so M is
-prime over N
: J. We should
check that N
: a
: I) it is also an
-decomposition of M
2
above
_
B
A
_
, i.e.
Denition 1.11(1),(2). Now in 1.11(1), clauses (a)-(h) are immediate, so let us
check clause (i) (in 1.11(2)). Let I), now is a
: ) I really
maximal (among independent over N
sets of elements of M
2
realizing a type from
P
= p S(N
) : p orthogonal to N
)?. This should be clear from clause (a)

(and basic properties of dependencies and regular types).
2) By 1.14(5).
3) Left to the reader.
1.24
1.25 Conclusion: Assume M
1
M
2
are
-saturated and
_
B
1
A
1
_

_
B
2
A
2
_
both in
(M
1
). If clause (a) (equivalently (b) or (c)) of 1.24 holds for
_
B
1
A
1
_
, M
1
, M
2
then
they hold for
_
B
2
A
2
_
, M
1
, M
2
.
Proof. By 1.24(2), clause (a) for
_
B
1
A
1
_
, M
1
, M
2
implies clause (a) for
_
B
2
A
2
_
, M
1
, M
2
.
1.25
(
4
0
1
)

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

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

32 SAHARON SHELAH
1.26 Claim. If
_
B
1
A
1
_
(M) and N
, a
: I) is an
-decomposition of M
above
_
B
1
A
1
_
and M
M is
-saturated and
_
I
N
and is an ordinal
then
tp
_
_
B
1
A
1
_
, M
= tp
_
_
B
1
A
1
_
, M
Proof. We prove this by induction on (for all B, A, N
, a
: I), I, M and M
as above). We can nd an
-decomposition N
, a
: J) of M with I J (by
1.13(4) + 1.13(2)) such that JI ,= ) and 0) and so M is
-prime
over
_
J
N
and also over M
: JI.
Case 0: = 0.
Trivial.
Case 1: is a limit ordinal.
Trivial by induction hypothesis (and the denition of tp
).
Case 2: = + 1.
We can nd M
which is
-prime over
_
I
N
, so as equality is transitive
it is enough to prove
tp
__
B
1
A
1
_
, M
_
= tp
__
B
1
A
1
_
, M
_
and
tp
__
B
1
A
1
_
, M
_
= tp
__
B
1
A
1
_
, M
_
.
By symmetry, this means that it is enough to prove the statement when M
is
-prime over
_
I
N
.
Looking at the denition of tp
+1
and remembering the induction hypothesis
our problems are as follows:
First component of tp
:
(
4
0
1
)

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

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

given
_
B
1
A
1
_

a
_
B
2
A
2
_
, B
2
M, it suces to nd
_
B
3
A
3
_
such that:
() there is f AUT(C) such that: f B
1
= id
B
1
, f(A
2
) = A
3
,
f(B
2
) = B
3
and B
3
M
and tp
[
_
B
2
A
2
_
, M] = tp
[
_
B
3
A
3
_
, M
]
(pedantically we should replace B
, A
by indexed sets).
We can nd J
, M
such that:
(i) I J
J, [J
I[ <
0
, J
closed under initial segments,

(ii) M
M is
-prime over M
: J
I
(iii) B
2
M
.
The induction hypothesis for applies, and gives
tp
_
_
B
2
A
2
_
, M
= tp
_
_
B
2
A
2
_
, M
.
By 1.14(4) there is g, an isomorphism from M
onto M
such that g B
1
= id.
So clearly g(B
2
) M
hence
tp
_
_
B
2
A
2
_
, M
= tp
_
_
g(B
2
)
g(A
2
)
_
, M
.
So
_
B
3
A
3
_
=: g
_
A
2
B
2
_
is as required.
Second component for tp
:
So we are given , a tp
type, (and we assign the lower part as B) and we have to

prove that the dimension in M and in M
are the same, i.e.

dim(I, M) = dim(I
, M), where: I = c M : = tp
__
c
B
1
_
, M
_
and
I
=
_
c M
: = tp
__
c
B
1
_
, M
__
.
Let p be such that: tp
_
_
c
B
1
_
, M
_
= p =
c
B
1
. Necessarily p A
1
and p is
regular (and stationary).
Clearly I
I, so without loss of generality I ,= hence p is really well dened,

now
() for every c I for some k < , c
realizing p for < k we have c

depends on c
0
, c
1
, . . . , c
k1
over B
1
.
[Why? Clearly p N
<>
(as B
1
A
1
N
<>
and p A
1
) hence
tp
(
_
J\I
N
, N
<>
) p hence
(
4
0
1
)

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

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

34 SAHARON SHELAH
tp
(
_
J\I
N
, M
) p, but M is
-prime over M
_
J\I
N
hence
by [Sh:c, V,3.2,p.250] for no c MM
is tp(c, M
) a stationarization of
p hence by [Sh:c, V,1.16](3) clearly () follows].
If the type p has depth zero, then (by 1.14(7)):
I = c M : tp(c, B) = p and
I
= c M
: tp(c, B) = p.
Now we have to prove dim(I, A) = dim(I
, A), as A is -nite and M, M
are
-saturated and I
I clearly
0
dim(I
, A) dim(I, A). Now the equality

follows by () above.
So we can assume p has depth > zero, hence (by [Sh:c, X,7.2]) that the type
p is trivial; hence, see [Sh:c, X7.3], in () without loss of generality k = 1 and
dependency is an equivalence relation, so for same dimension it suces to prove
that every equivalence class (in M i.e. in I) is representable in M
i.e. in I
. By
the remark on () in the previous sentence (d
1
I)(d
2
I
)
_
d
1
B
1
d
2
. So it
is enough to prove that:
if d
1
, d
2
M realize same type over B
1
, which is (stationary and) regular,
and are dependent over B
1
and d
1
M
then there is d
2
M
such that
d
2
B
1
+d
1
=
d
2
B
1
+d
1
and tp
__
B
1
+d
2
B
1
_
, M
= tp
__
B
1
+d
2
B
1
_
, M
.
Let M
0
= N
. There are J
, M
1
, M
+
1
such that
()
1
(i) J
J is nite (and of course closed under initial segments)

(ii) ) J
, 0) / J
(iii) M
1
M is
-prime over N
: J
(iv) M
+
1
M is
-prime over M
1
M
(and M
1
M
0
M
)
(v) d
2
M
+
1
.
Now the triple
_
B
1
+d
2
B
1
_
, M
1
, M satises the demand on
_
B
1
A
1
_
, M
, M (because
_
B
1
A
1
_

_
B
1
+d
2
B
1
_
,by 1.25. Hence by the induction hypothesis we know that
tp
_
_
B
1
+ d
2
B
1
_
, M
= tp
_
_
B
1
+ d
2
B
1
_
, M
+
1
.
(
4
0
1
)

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

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

By 1.29(4) there is an isomorphism f from M
+
1
onto M
which is the identity on

B
1
+ d
1
; let d
2
= f(d
2
) so:
tp
_
_
B
1
+ d
2
B
1
_
, M
+
1
= tp
_
_
B
1
+ d
2
B
1
_
, M
.
Together
tp
_
_
B
1
+ d
2
B
1
_
, M
= tp
_
_
B
1
+ d
2
B
1
_
, M
.
As d
1
, d
2
is not independent over B
1
, also f(d
1
), f(d
2
) = d
1
, f(d
2
) is not
independent over B
1
, hence, as p is regular
() d
2
, f(d
2
) is not independent over B
2
.
Together we have proved
, hence nished proving the equality of the second

component.
Third component: Trivial.
So we have nished the induction step, hence the proof.
1.26
1.27 Claim. 1) Suppose M is
-saturated, A B M,
_
B
A
_
,
2
=1
[A A

M],
A = ac(A), A
are -nite,
A
1
A
=
A
2
A
, B
A
A
1
and B
A
A
2
.
Then tp
__
A
1
B
A
1
_
, M
= tp
__
A
2
B
A
2
_
, M
for any ordinal .

2) Suppose M is
-saturated, B M,
_
B
A
_
,
2
=1
[A A
M], A = ac(A),
B = ac(B), A
= ac(A
), A
is -nite,
A
1
A
=
A
2
A
, B
A
A
1
, B
A
A
2
, f : A
1
onto
A
2
an elementary mapping, f A = id
A
, g f id
B
, g elementary mapping from
B
1
= ac(B A
1
) onto B
2
= ac(B A
2
).
Then g
_
tp
__
B
1
A
1
_
, M
_
= tp
__
B
2
A
2
_
, M
for any ordinal .

3) Assume that
(a) A
= acl(A
) B
= acl(B
) M
for = 1, 2
(b) A
A
+
acl(A
+
) M
for = 1, 2
(
4
0
1
)

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

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

36 SAHARON SHELAH
(c) B
A
+
for = 1, 2
(d) f is an elementary mapping from A
1
onto A
2
(e) g is an elementary mapping from A
+
1
onto A
+
2
(f) f A
1
= g A
1
(g) h is an elementary mapping from B
+
1
= acl(B
1
A
+
1
) onto B
+
2
= acl(B
2
A
+
2
) extending f and g
(h) f(tp
[
_
B
1
A
1
_
, M
1
]) = tp
[
_
B
2
A
2
_
, M
2
].
Then h(tp
[
_
B
+
1
A
+
1
_
, M
1
]) = tp
[
_
B
+
2
A
+
2
_
, M
2
].
Proof. 1) Follows from part (2).
2) We can nd A
3
M such that:
(i)
A
3
A
=
A
1
A
(ii) A
3
A
(B A
1
A
2
).
Hence without loss of generality A
1
B
A
2
and even
A
B, A
1
, A
2
. Now we can nd
N
<>
, an
-prime model over , N

<>
M, A N
<>
and (B A
1
A
2
)
A
N
<>
(e.g. choose A
1
A
i
B
: M indiscernible over A, A
1
= A
1
, A
2
=
A
2
, B
= B and let N
<>
M be
-primary over
_
n<
(A
n
1
A
n
2
B
n
A)).
Now nd N
, a
: J) an
-decomposition of M with
dcl(a
<0>
) = dcl(B), dcl(a
<1>
) = dcl(A
1
), dcl(a
<2>
) = dcl(A
2
).
Let I = J : =<> or < 0 > and J
= I < 1 >, < 2 >. Let

N
2
<>
M
be
-prime over N
<1>
N
<2>
. By 1.21 there is N
2
, a
: I) an
-
decomposition of M above
_
B
A
_
such that N
, a
: I)
direct
N
2
, a
: I).
Let M
M be
-prime over
_
I
N
2
and M
be
-prime over
_
I
N
. So
M
M and M
is
-prime over M
N
<1>
N
<2>
.
Now by 1.26 we have tp
__
B
A
_
, M
= tp
__
B
A
_
, M
for = 1, 2 hence it suces

to nd an automorphism of M
extending g. Let B
+
= ac(N
<>
B), A
= ac(B
A
), let a
list A
be such that a
2
= g(a
1
). Clearly tp(a
, B
+
A B and ac(B) = B and so stp(a
1
, B
+
) = stp(a
2
, B
+
). Also tp
(A
2
, B
+
A
1
)
does not fork over A hence tp(a
2
, B
+
a
1
) does not fork over A B
+
hence
a
1
, a
2
is independent over B
+
hence there is an elementary mapping g
+
from
(
4
0
1
)

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

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

ac(B
+
a
1
) onto ac(B
+
a
2
), g
+
id
B
+ g and even g
= g
+
(g
+
)
1
is an
elementary embedding.
Let a
1
lists ac(N
<>
A
1
) so clearly a
2
=: g
+
(a
1
) list ac(N
<>
A
2
). Clearly
g
(a
1
a
2
) is an elementary mapping from a
1
a
2
onto itself. Now N
2
<>
is
-
primary over N
<>
A
1
A
2
and N
<>
A
1
A
2
a
1
a
2
ac(N
<>
A
1
A
2
) so by
1.18(10) N
2
<>
is
-primary over N
<>
a
1
a
2
hence we can extend g
(a
1
a
2
)
to an automorphism h
<>
of N
2
<>
so clearly h
<>
N
<>
= id
N
<>
. Let a
+
1
list
ac(B
+
A
1
) and a
+
2
= g
+
(a
+
1
). So tp(a
+
, N
2
<>
) does not fork over a
1
( N
2
<>
)
and ac(a
1
) = Rang(a
1
) (= ac(N
<>
A
1
)) and h
<>
a
1
= g
+
a
1
hence h
<>
g
+
is an elementary embedding. Remember g
+
is the identity on B
+
= ac(N
<>
B),
and tp
(N
<0>
, N
2
<>
) does not fork over N
<>
hence tp
(N
<0>
, B
+
N
2
<>
) does not
fork over B
+
, so as ac(B
+
) = B
+
necessarily (h
<>
g
+
) id
N
<0>
is an elementary
embedding. But this mapping has domain and range including N
<0>
N
2
<>
and
included in N
2
<0>
, but the latter is
-primary and
-minimal over the former.

Hence (h
<>
g
+
) id
N
<>
can be extended to an automorphism of N
2
<0>
which
we call h
<0>
.
Now we dene by induction on n [2, ) for every I of length n, an
automorphism h
of N
2
extending h
id
N
, which exists as N
2
is
-primary over
N
2
(and N
2
). Now
_
I
h
is an elementary mapping (as N

2
: I)
is a non-forking tree; i.e. 1.13(10)), with domain and range
_
I
N
2
hence can be
extended to an automorphism h
of M
, (we can demand h
= id
M
but
not necessarily). So as h
extends g, the conclusion follows.

3) Similarly to (2).
1.27
1.28 Claim. 1) For every = tp
__
B
A
_
, M
, and a,
b listing A, B respectively
there is = ( x
A
, x
B
) L
,
(q.d.) of depth such that:

tp
_
_
B
A
_
, M
= M [= [a,
b].
2) Assume
M
1
,M
2
of 1.4 holds as exemplied by the family F and
_
B
A
_
(M
1
)
and g F, Dom(g) = B; and an ordinal then
tp
__
B
A
_
, M
_
= tp
__
g(B)
g(A)
_
, M
2
_
.
3) Similarly for tp
([A], M), tp
[M].
(
4
0
1
)

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

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

38 SAHARON SHELAH
Proof. Straightforward (remember we assume that every rst order formula is
equivalent to a predicate).
1.28
1.29 Proof of Theorem 1.2. [The proof does not require that the M
are
-
saturated, but only that 1.27, 1.28 hold except in constructing g
()
(see
14
,
15
in 1.30(E), we could instead use NOTOP].
So suppose
()
0
M
1
L
,
(d.q.)
M
2
or (at least)
M
1
,M
2
from 1.4 holds.
We shall prove M
1

= M
2
. By 1.28 (i.e. by 1.28(1) if the rst possibility in ()
0
holds and by 1.28(2) if the second possibility in ()
0
holds)
()
1
tp
[M
1
] = tp
[M
2
].
So it suces to prove:
1.30 Claim. Assume that T is countable. If M
1
, M
2
are
-saturated models (of

T, T as in 1.5), then:
()
1
M
1

= M
2
.
Proof. Let W
k
, W
k
: k < ) be a partition of to innite sets (so pairwise disjoint).
1.31 Explanation: (If seems opaque, the reader may return to it after reading parts
of the proof).
We shall now dene an approximation to a decomposition. That is we are ap-
proximating a non-forking tree N
, a
: I
) of countable elementary submodels

of M
for = 1, 2 and f
: I
) such that f
an isomorphism from N
1
onto N
2
increasing with such that M
is
-prime over
_
I
.
In the approximation Y we have:
() I approximating I
,
[it will not be I
n
Ord but we may discover more immediate successors
to each I; as the approximation to N
improves we have more regular

types, but some member of I will be later will drop]
() A
approximates N
and is -nite
() a
is the a
(if survives, i.e. will not be dropped)

(
4
0
1
)

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

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

() B
, b
,m
expresses commitments on constructing A
: we promise
B
and B
is countable; b
,m
for m < list B
(so in the choice

B
there is some arbitrariness)

() f
approximate f
() p
,m
also expresses commitments on the construction.
Since there are innitely many commitments that we must meet in a construction
of length and we would like many chances to meet each of them, the sets W
k
, W
k
are introduced as a further bookkeeping device. At stage n in the construction we
will deal e.g. with the b
,m
for that are appropriate and for m W
k
for some
k < n and analogously for p
,m
and the W
k
.
Note that while the A
satisfy the independence properties of a decomposition,

the B
do not and may well intersect non-trivially. Nevertheless, a conict arises

if an a

falls into B
since the a

are supposed to represent independent
elements realizing regular types over the model approximated by A
but now a

is in that model. This problem is addressed by pruning from the tree I.
1.32 Denition. An approximation Y to an isomorphism consist of:
(a) natural numbers n, k
and index set: I

n
Ord
(and n minimal)
(b) A
, B
, a
, b
,m
: I and m
_
k<k
W
k
) for = 1, 2 (this is an approxi-
mated decomposition)
(c) f
: I)
(d) p
,m
: I and m
_
k<k
k
)
such that:
(1) I closed under initial segments
(2) <> I
(3) A
, A
is -nite, ac(A
) = A
, B
is countable,
B
= b
,m
: m
_
k<k
W
k
(4) A
if I
(
4
0
1
)

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

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

40 SAHARON SHELAH
(5) if I<>, then
a
)
is a (stationary) regular type and a
; if in
addition g() > 1 then
a
)
A
)
(note that we may decide a
<>
be
not dened or A
<>
)
(6)
A
+a
a
A
if I, g() > 0
(7) if I, not -maximal in I, then the set a
: I and
= is a
maximal family of elements realizing over A
regular types A
)
(when
is dened), independent over

_
A
, B
_
, (and we can add: if
1
=
2
= and
a
1
A
2
A
then a
1
/A
= a
2
/A
)
(8) f
is an elementary map from A

1
onto A
2
(9) f
(
)
f
when I, g() > 0

(10) f
(a
1
) = a
2
(11) () f
_
tp
_
_
A
1
A
1
(
)
_
, M
1
__
= tp
_
_
A
2
A
2
(
)
_
, M
2
_
when I<>
() f
<>
_
tp
_
A
1
<>
, M
1
_
= tp
_
A
2
<>
, M
2
(12) B
moreover, B

na
M
, i.e., if a N
, b M
and M
[=
(b, a) there for some b
, [= (b
, a) and b / acl( a) b
/ acl(A)
(13) p
,m
: m
_
k<k
k
) is a sequence of types over A
(so Dom(p
,m
) may be
a proper subset of A
).
1.33 Notation. We write n = n
Y
= n[Y ], I = I
Y
= I[Y ], A
= A
[Y ],
B
= B
[Y ], f
= f
Y
= f
[Y ], a
= a
[Y ], b
= b
[Y ], k
= k
Y
= k
[Y ] and
p
,m
= p
,k
[Y ].
Remark. We may decide to demand: each
a

A
is strongly regular; also: if two

such types are not orthogonal then they are equal (or at least have same witness
for (,
a

A
) regular). This is easy here as the models are
-saturated (so take

p
p, rk(p
) minimal).
1.34 Observation. ()
1
implies that there is an approximation, (see 1.29).
(
4
0
1
)

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

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

Proof. Let I = <>, A
<>
= ac(), k
= 1 and then choose countable B
<>
to
satisfy condition (12) and then choose f
, p
k
, b
,m
(for k W
0
and m W
0
) as
required.
1.35 Main Fact. For any approximation Y, i
_
k<k
Y
(W
k
W
k
) and m n
Y
and
() 1, 2 we can nd an approximation Z such that:
(
)() n
Z
= Maxm + 1, n
Y
, I
Z

m
Ord = I
Y

m
Ord,
(we mean m not n
Y
) and k
Z
= k
Y
+ 1
() (a) if I
Y
, g() < m then
A
[Z] = A
[Z],
a
[Z] = a
[Z]
B
[Z] = B
[Y ]
(b) if I
Y
I
Z
, k < k
Y
and j W
k
then
p
,j
[Z] = p
,j
[Y ]
(c) if I
Y
I
Z
, k < k
Y
and j W
k
then
b
,j
[Z] = b
,j
[Y ]
()
1
if I
Y
, g() = m, k < k
Y
and
4
i W
k
and the element b M
()
satises clauses (a), (b) below then for some such b we have: A
()
[Z] =
ac(A
()
[Y ] b); where
(a) b
()
,i
[Y ] / A
()
[Y ] and g() > 0

b
A
()
[Y ]
a
A
()
[Y ]
(b) one of the conditions (i),(ii) listed below holds for b
(i) b = b
()
,i
[Y ] and
g() > 0
b
A
()
[Y ]
a
A
()
[Y ] or
(ii) for no b is (i) satised (so g() > 0) and b M
()
,
b
,i
A
()
[Y ]
b and g() > 0
b
A
()
[Y ]
a
A
()
[Y ]
4
recall that i is part of the information given in the main fact, and of course, k is uniquely
determined by i
(
4
0
1
)

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

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

42 SAHARON SHELAH
()
2
assume I
Y
, g() = m, k < k
Y
and i W
k
then we have:
(a) if p
()
,i
is realized by some b M
()
such that
Rk
_
b
A
()
[Y ]
, L,
_
= R
_
p
()
,i
, L,
_
and
_
g() > 0
b
A
()
[Y ]
a
A
()
[Y ]
_
then for some such b we have
A
()
[Z] = ac
_
A
()
[Y ] b
_
(b) if the assumption of clause (a) fails but p
()
,i
is realized
by some b M
()
A
()
such that
_
g() > 0
b
A
()
[Y ]
a
A
()
[Y ]
_
then for some such b we have
A
()
[Z] = ac
_
A
()
[Y ] b
_
() If I
Y
and g() = m, then B
[Z] = b
,j
[Y ] : j W
k
: k < k
is a countable subset of M
, containing B
[Z] : and Y B
[Y ],
with B
[Z] M
moreover B
[Z]
na
M
i.e. if a B
[Z], (x, y) is
rst order and (x M
ac( a)) (x, a) then (x B
[Z]ac( a))( x, a))

and a
<>
[Y ] : ) I
Y
and a
<>
[Y ] / B
[Z] is independent over

(B
[Z], A
[Y ])
() if I
Y
, g() > m, then I
Z
a
(m+1)
[Y ] / B
m
[Z]
() if I
Y
I
Z
, g() > m then A
[Z] = ac(A
[Y ] A
m
[Z]) and
B
[Z] = B
[Y ]
() if I
Z
I
Y
then
I
Y
and g() = m + 1
() p
,i
[Z] : i W
Z
1
is rich enough, e.g. include all nite types over A
() b
,i
: i W
k
z
1
list B
[Z], each appearing innitely often.

(
4
0
1
)

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

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

Proof. First we choose A
()
[Z] for I of length m according to condition

() = ()
1
+()
2
. (Note: one of the clauses ()
1
, ()
2
necessarily holds trivially as
_
k
W
k

_
k
W
k
= ).
Second, we choose (for such ) an elementary mapping f
Z
extending f
Y
and a set
A
3()
[Z] M
3()
satisfying f
Z
is from A
1
[Z] onto A
3()
[Z] such that

()
2
if m > 0, then
f
Z
_
tp
__
A
1
[Z]
A
1
[Y ]
_
, M
1
__
= tp
__
A
2
[Z]
A
2
[Y ]
_
, M
2
_
()
3
if m = 0, then f
Z
_
tp
_
A
1
[Z], M
1
__
= tp
_
A
2
[Z], M
2
_
.
[Why possible? If we ask just the equality of tp
for an ordinal , this follows by

the rst component of tp
+1
. But (overshooting) for
_
(|M
1
| +|M
2
|)
|T|
+
,
equality of tp
implies equality of tp
].
Third, we choose B
[Z] for I
Y
, g() = m according to condition () (here we
use the countability of the language, you can do it by extending it times) in both
sides, i.e. for = 1, 2.
Fourth, let I
= I : if g() > m then a
(m+1)
[Y ] / B
m
[Z] (this will be
I
Y
I
Z
).
Fifth, we choose A
[Z] for I
: if g() < m, let A
[Z] = A
[Y ], if g() = m
this was done, lastly if g() > m, let A
[Z] = ac
_
A
[Y ] A
m
[Z]
_
.
Sixth, by induction on k n
Y
we choose f
Z
for I
of length k: if g() < m,

let f
Z
= f
Y
, if g() = m this was done, lastly if g() > m choose an elementary

mapping from A
1
onto A
2
extending f
Y
f
Z
(possible as f
Y
f
2
is an elementary
mapping and Dom(f
Y
) Dom(f
Z
) = A
()
, Dom(f
Y
A
()
Dom(f
Z
)
and A
()
= ac(A
()
).) Now f
Z
satises clause (11) of Denition 1.32 when

g() > m by applying 1.27(3).
Seventh, for I
, of length m < n
Z
, let v
=: : ) I, and we choose
a
1
<>
[Z] : u
, [ u
) / I], a set of elements of M

1
realizing
(stationary) regular types over A
1
[Z], orthogonal to A
[Y ] when g() > 0, such

that it is independent over
_
a
1
<>
[Y ] : ) I
B
1
[Z], A
1
[Z]
_
and max-
imal under those restrictions. Without loss of generality sup(v
) < min(u
) and
for
1
v
and
2
u
we have:
(
4
0
1
)

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

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

44 SAHARON SHELAH
()
1
if (for the given
2
and )
1
is minimal such that
a
1
<
1
>
[Z]
A
1
[Z]

a
1
<
2
>
[Z]
A
1
[Z]
then
a
1
<
1
>
[Z]
A
1
[Z]
=
a
1
<
2
>
[Z]
A
1
[Z]
()
2
if
1
<
2
1
and a
1
[Z]/A
1
[Z] = a
1
[Z]/A
1
[Z] and for some b M

1
realizing
a
1
<
2
>
[Z]
A
1
[Z]
we have
b
A
1
[Z]
a
1
<
2
>
and
tp
_
_
b
A
1
<
2
>
_
, M
_
= tp
__
a
1
<
1
>
A
1
<
2
>
_
, M
_
and
1
is minimal (for the given
2
and ) then
tp
_
_
a
1
<
2
>
A
1
<
2
>
_
, M
_
= tp
_
_
a
1
<
1
>
A
1
<
1
>
_
, M
_
.
Easily (as in [Sh:c, X]) if u
and ) I
then
a
1
<>
[Z]
A
1
[Z]

a
1
<>
[Y ]
A
1
[Y ]
.
For u
let A
1
<>
[Z] = ac
_
A
1
[Y ] a
1
<>
[Z]
_
.
Eighth, by the second component in the denition of tp
+1
(see Denition 1.9) we
can choose (for u
) a
2
<>
[Z], A
2
<>
[Z] and then f
Z
<>
as required (see (7)
of Denition 1.32).
Ninth and lastly, we let I
Z
= I
< >: I
, g() = m < n
Z
and u
and we choose B
for I
Z
I
Y
and the p
,i
, b
,j
as required (also in other case
left).
1.35
1.36 Finishing the Proof of 1.11. We dene by induction on n < an approximation
Y
n
= Y (n). Let Y
0
be the trivial one (as in observation 1.30(C)).
Y
n+1
is gotten from Y
n
as in 1.35 for m
n
, i
n
n,
n
() 1, 2 dened by
reasonable bookkeeping (so i
n

_
k<k
Y (n)
(W
k
W
k
)) such that any triples appear
innitely often; without loss of generality: if n
1
< n
2
& I
n
1
I
n
2
then

n
2
n=n
1
I
n
.
Let I
= I[] = lim(I
Y (n)
) =: : for every large enough n, I

n
;
for I
let: A
[] =
_
n<
A
[Y
n
], f
[] =
_
n<
f
Y (n)
and
(
4
0
1
)

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

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

B
[] =
_
n I
and I
>
Ord is closed under initial segments
1
for I
, B
[Y
n
] : n < and I[Y
n
]) is an increasing sequence of
na
-elementary submodels of M
[Why? By clause (12) of Denition 1.32, Main Fact 1.35, clauses ()(a), (), ().]
hence
2
for I
, B
[]
na
M
.
Also
3
I
[] B
[].
[Why? Because for innitely many n, m
n
= g() and clause () of Main Fact 1.35].
4
if I[Y
n
1
] I
= and n
1
n
2
then
A
[Y
n
1
]
[Y
n
1
]
A
[Y
n
2
].
[Why? Prove by induction on n
2
(using the non-forking calculus), for n
2
= n
1
this
is trivial, so assume n
2
> n
1
. If m
(n
2
1)
> g() we have A
[Y
n
2
] = A
[Y
n
2
1
] (see
1.35, clause ()(a) and we have nothing to prove. If m
(n
2
1)
< g() then we note
that A
[Y
n
2
] = acl(A
[Y
n
2
1
]A
m
(n
2
1)
[Y
n
2
]) and A
[Y
n
2
1
]
m
(n
2
1)
A
m
(n
2
1)
[Y
n
2
]
(as I[Y
n
2
], by 1.35 clause () last phrase) and now use clauses (5), (6) of De-
nition 1.35. Lastly if m
(n
2
1)
= g() again use I[Y
n
2
] by 1.35, clause (), last
phrase].
5
if I[Y
n
1
] I
= and n
1
n
2
then
A
[]
A
[] + a
[]

a
A
.
[Why? By clause (6) of Denition 1.32, and orthogonality calculus].
6
if I
, then A
[] B
[] M
moreover
7
A
()
[]
na
B
[]
na
M
.
(
4
0
1
)

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

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

46 SAHARON SHELAH
[Why? The second relation holds by
2
. The rst relation we prove by induction
on g(); clearly A
[] = ac(A
[]) because A
[Y
n
] increases with n by 1.35 and
A
[Y
n
] = acl(A
[Y
n
]) by clause (3) of Denition 1.32. We prove A
()
[]
na
B
[] by induction on m = g(), so suppose this is true for every m
< m, m =
g(), I
, let (x) be a formula with parameters in A
[] realized in M
as
above say by b M
. As A
[Y
n
] : n < , Y
n
) is increasing with union A
[],
clearly for some n we have b
[Y
n
]
A
[].
So (x) = p
,i
for some i and for some n
> n dening Y
n
+1
we have used
1.35 with ((), i, m) there being (, i, g()) here, hence we consider clause ()
2
of
1.35. So the case left is when the assumption of both clauses (a) and (b) of ()
2
fail, so we have g() > 0 and
b
/ A
[Y
n
], b
[= [b
]
b
[Y
n
]
A
[Y
n
].
We can now use the induction hypothesis (and [BeSh 307, 5.3,p.292]).]
8
if I
and = 1, 2, then
a
<>
[] : ) I
is a maximal subset of
c M
:
c
A
[]
regular, c
[]
B
[] and g() > 0

c
A
[]
A
[]
independent over (A
[], B
[]).
[Why? Note clause (7) of Denition 1.32 and clause () of Main Fact 1.35].
9
A
<>
[] = B
<>
[]
[Why? By the bookkeeping every b B
<>
[] is considered for addition to A
<>
[]
see 1.35, clause ()
1
, subclause (b)(i) and for ) there is nothing to stop us].
10
if I
<> and p S(A
[]) is regular orthogonal to A
[] then
B
[]
A
[]

p.
[Why? If not, as A
[]
na
B
[] by [BeSh 307, Th.B,p.277] there is

c B
[]A
[] such that:
c
A
[]
is p. As c B
[] =
_
n<
B
[Y
n
], for every n <
large enough c B
[Y
n
], and p does not fork over A
[Y
n
]. So for some such n
(
4
0
1
)

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

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

the triple (i
n
,
n
, m
n
) is such that
n
= , m
n
= g() and b
,i
n
= c, so by clause
()
1
(b)(ii) of 1.35 we have c A
[Y
n
] A
[].]
11
if I
, 1, 2 then a
<>
: ) I
is a maximal subset of
c M
:
c
A
[]
regular, A
[] when meaningful independent over

A
[].
[Why? If not, then for some c M, a
: ) I
c is indepen-
dent over A
[] and tp(c, A
[]) is regular (and stationary). Hence by

10
we
have a
[Y
n
] : ) I
c is independent over (A
[], B
[]). Now for

large enough n we have c
[Y
n
]
A
[] and by
10
we have c
[Y
n
]
B
[], hence
c
[]
B
[Y
n
], and so by Denition 1.32(7) c a
[Y
n
] : ) I[Y
n
] is
not independent over (A
[Y
n
], B
[Y
n
]), but a
[Y
n
] : ) I[Y
n
] is inde-
pendent over (A
[Y
n
], B
[Y
n
]). So there is a nite set w of ordinals such that
w ) I[Y
n
] and c a
[Y
n
] : w is not independent over
(A
[Y
n
], B
[Y
n
]), and without loss of generality w is minimal. Let n
1
[n, ) be
such that w & a
<>
B
[] a
[Y
n
1
]; clearly exist as w is nite
and let u = w : a
<>
/ B
[]; clearly u < > I
. Now
a
[] : ) I
[] includes a
[Y
n
] : w, easy contradiction to
the second sentence above.]
12
f
=
_
m<
f
[Y
n
] (for I
) is an elementary map from A

1
[] onto A
2
[].
[Easy].
13
f
=:
_
I
is an elementary mapping from

_
I
A
1
[] onto
_
I
A
2
[].
[Clear using by
5
+
6
+
12
and non-forking calculus].
14
We can nd d
: < ()) such that:

(a) d
, < d
,= d
tp
_
d
,
_
I[]
A
[] d
: <
_
is
-isolated and F
0
-isolated, and
(b) g
=
_
I
(d
1
, d
2
) : < ()) is an elementary mapping,

(
4
0
1
)

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

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

48 SAHARON SHELAH
(c) () is maximal, i.e., we cannot nd d
1
()
such that the demand in
(a) holds for () + 1.
[Why? We can try to choose by induction on , a member d
1
of M
1
_
I[]
d
1
:
< such that tp(d
1
,
_
I[]
A
[] d
1
: < ) is
-isolated and F
0
-isolated.
So for some (), d
1
is well dened i < () (as < d

1
,= d
1
M
1
).
Now choose by induction on < (), d
2
M
2
as required above, possible by M
2
i
being
-saturated (see [Sh:c, XII,2.1,p.591], [Sh:c, IV,3.10,p.179].]
15
Dom(g
()
), Rang(g
()
) are universes of elementary submodels of M
1
, M
2
respectively called M
1
, M
2
respectively.
[Why? See [Sh:c, XII,1.2](2),p.591 and the proof of
14
.
Alternatively, choose a formula (x, a) such that:
(a) a Dom(g
()
) and [= x(x, a) but no b Dom(g
()
) satisfy (x, a)
(b) under clause (a), Rk((x, a), L
|T|
, ) is minimal
(or just has no extension in S(Dom(g
()
)) forking over a).
Let
(x, y
) : < list that L

(T)
-formulas and we choose by induction on as
formula
n
(x, a
n
) such that:
(i) a Dom(g
()
)
(ii) [= (x)
n
(x, a
n
)
(iii)
n+1
(x, a
n+1
)
n
(x, a
n
)
(iv)
0
(x, a
0
) = (x, a)
(v) for any formula
(x, a
) satisfying the demands on

n+1
(x, a
n+1
) we have
Rk(
n+1
(x, a
n+1
),
n
(x, y
n
), 2) < Rk(
(x, a),
n
(x, y), 2)
(on this rank see [Sh:c, II,2]).
So p =
n
(x, a
n
) : n < has an extension in S(Dom(g
()
)) call it q. Now q
is
-isolated because (x, a) q S(Dom(g

()
). For every n,
n+1
(x, a
n
) q
n
(x, y
n
) by clause (v) above so as
n+1
(x, a
n
) q and this holds for every n
clearly q is F
0
-isolated.
16
If M
,= M
then for some d M
,
d
M
is regular.
(
4
0
1
)

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

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

[Why? By [BeSh 307, Th.5.9,p.298] as N

na
M
by
7
].
17
if M
,= M
then for some I
, there is d M
such that
d
A
[]
is
regular, d
[]
M
and
_
g() > 0
d
A
[]
A
[]
_
.
[Why? By [Sh:c, XII,1.4,p.529] every non-algebraic p S(M
) is not orthogonal
to some A
[] so by
16
we can choose I
and d M
such that
d
M
is
regular A
[]; without loss of generality g() is minimal, now A
[]
na
M
and
by [BeSh 307, 4.5,p.290] without loss of generality d
[]
M
; the last clause is by

g() minimal].
18
M
= M
.
[Why? By
11
+
17
].
19
there is an isomorphism from M
1
onto M
2
extending
_
I
.
[Why? By
14
+
15
we have M
1

= M
2
, so by
18
we are done].
1.36

1.30
1.37 Lemma. Assume B
A
C, A = ac(A) = B C and A, B, C are -nite,
A B C M, M an
-saturated model of T. For notational simplicity make A

a set of individual constants.
Then tp
L
,
(d.q.)
(B + C; M) = tp
L
,
(d.q.)
(B; M) + tp
L
,
(d.q.)
[C; M]
where
1.38 Denition. 1) For any logic L and

b a sequence from a model M, let
tp
L
(
b; M) =
_
( x) :M [= [B], a formula in the vocabulary of M,
from the logic L (with free variables from
x, where x = x
i
: i < g(
b)))
_
.
2) Replacing

b by a set B means we use the variables x
b
: b B).
3) Saying p
1
= p
2
+p
3
in 1.37 means that we can compute p
1
from p
2
and p
3
(and
the knowledge how the variables t and the knowledge of T, of course).
(
4
0
1
)

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

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

50 SAHARON SHELAH
Proof of the Lemma 1.37.
It is enough to prove:
1.39 Claim. Assume
(a) M
1
, M
2
are
-saturated and
(b) A
i
1
A
i
0
A
i
2
for i = 1, 2
(c) A
i
0
= ac(A
i
0
) and A
i
m
is -nite for i = 1, 2 and m < 3
(d) for m = 0, 1, 2 we have f
m
: A
1
m
onto
A
2
m
is an elementary mapping preserv-
ing tp
(in M
1
, M
2
respectively) and
(e) f
0
f
1
, f
2
.
Then there is an isomorphism from M
1
onto M
2
extending f
1
f
2
.
Proof of 1.39. Repeat the proof of 1.5, but starting with Y
0
such that
A
<>
[Y
0
] = A
0
, A
<>
[Y
0
] = A
1
, A
<1>
[Y
0
] = A
2
, f
Y
0
<>
= f
0
, f
Y
0
<0>
= f
1
, f
Y
0
<1>
= f
2
and
that ), 0), 1) belongs to all I[Y
0
]. During the construction we preserve 0), 1)
I[Y
n
] and for helping to preserve this we add also the demand
2,m
B
<>
[Y
n
]
0
A
1
A
2
.
During the proof, when we have to increase B
<>
, we use 1.18(1) + 1.16(1).
1.39
Discussion: A natural version of 1.39 to say is the conclusion only that tp
[
_
A
1
a
A
1
2
A
1
0
_
, M
1
] =
tp
[
_
A
2
1
A
2
2
A
2
0
_
, M
2
] and to prove this by induction on . The case = 0 and limit
are obvious. If = +1, for the condition of
a
, we use the induction hypothesis
and claim 1.27(1). The condition involving
b
is similar but harder.
1.39
(
4
0
1
)

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

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

2 Finer Types
We shall use here alternative types showing us probably a ner way to manipulate
tp.
2.1 Convention. T is superstable, NDOP
M, N are
-saturated C
eq
.
2.2 Denition.
3
=
_
_
b
a
_
: a
b are -nite
_
1
=
__
p
a
_
: a is -nite, p S( a) is regular (so stationary)
_
2
=
__
p, r
a
_
: a is -nite, p is a regular type of depth > 0,
p a (really only the equivalence class p/ matters),
r = r(x, y) S( a) is such that for (c,
b) realizing r,
c/( a +
b) is regular p, and
b
a
= (r y) p
_
.
We may add (to
x
) superscripts:
() f if a (or a
b) is nite
() s : for
3
if

b
a
is stationary, for
1
if p is stationary which holds always and
for
2
if r is stationary and every automorphism of C over a x p/
() c if a (or a,
b) are algebraically closed.

2.3 Claim. If p is regular of depth > 0 and p a and a is -nite then for some
a
, a a
ac( a) and for some q we have

_
p,q
a
_

s
2
.
Proof. Use, e.g., [Sh:c, V,4.11,p.272], assume

b
a
p; we can dene inductively equiv-
alence relations E
n
, with parameters from ac( a
), a
= a(
b/E
0
) (
b/E
n1
),
such that tp(
b/E
n
, ac( a
n
)) is semi-regular. By superstability this stop for some n
hence
b ac( a
n
). For some rst m tp(
b/E
m
, ac( a
n
)) is p, by [Sh:c, X,7.3](5),p.552
the type is regular (as because p is trivial having depth > 0; see [Sh:c, X,7.2,p.551]).
2.3
(
4
0
1
)

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

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

52 SAHARON SHELAH
2.4 Denition. We dene by induction on an ordinal the following (simultane-
ously): note if a denition of something depends on another which is not well
dened, neither is the something)
tp
1
_
_
p
a
_
, M
for
_
p
a
_

1
, a M
tp
2
_
_
p, r
a
_
, M
for
_
p, r
a
_

1
, a M
tp
3
_
_
b
a
_
, M
for
_
b
a
_

c
3
, a
b M
Case A = 0: tp
1
__
p
a
_
, M
is tp((c, a), ) for any c realizing p.

tp
2
_
_
p, r
a
_
, M
is tp((c,
b, a), ) for any (c,
b) realizing r
tp
3
_
_
b
a
_
, M
is tp((
b, a), )
(i.e., the type and the division of the variables between the sequences).
Case B = + 1:
(a) tp
1
__
p
a
_
, M
is:
Subcase a1: if p has depth zero, it is w
p
(M/ a) (the p-weight, equivalently, the
dimension)
Subcase a2: if p has depth > 0 (hence is trivial), then it is y,
y
a,p
) : y where
y
a,p
= dim(I
y
a,p
[M], a) where I
y
a,p
[M] =
_
c M : c realize p and y = tp
3
__
ac( a + c)
ac( a)
_
, M
_
where a
list acl( a) and c
list acl( a + c)
_
an alternative probably more transparent and simpler in use is:
(
4
0
1
)

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

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

y
a,p
= dim
_
c M : c realizes p and
y = tp
3
__
ac( a + c
)
ac( a)
_
, M
_
: c
p(M) and c
a
c
pedantically y = tp
3
_
< c
> a
a a
_
, M]; where
a
list acl( a) and

c
list acl( a + c
), c
p(M) and c
a
c
_
.
(b) tp
2
__
p,r
a
_
, M
is:
tp
1
_
_
c/
b
+
b
+
_
, M
_
for any (c,
b) realizing r,
b
+
= ac( a+
b), i.e.,

b
+
lists acl( a+
b) (so not well dened if we get at least two dierent cases; so remember
c/b
+
S(
b
+
)).
(c) tp
3
_
_
b
a
_
, M
_
is
_
p, tp
2
__
p,r
b
_
, M
) :
_
p,r
b
_

s
2
and p a
_
.
Case C: limit: For any 1, 2, 3 and suitable object OB:
tp
[OB, M] = tp
[OB, M] : < ).
2.5 Denition. 1) For
_
p
a
_

1
where a M, let (remembering 1.14(8)):
P
M
(
p
a
)
=
_
q S(M) : q regular and : q p or for some
c p(M) we have q P
M
(
c
a
)
_
.
2) For
_
p,r
a
_

2
let
P
M
(
p,r
a
)
=
_
q S(M) : q regular and : q p or for some
(c,
b) r(M), q P
M
(
c
a+
b
)
_
.
(
4
0
1
)

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

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

54 SAHARON SHELAH
3) For a set P of (stationary) regular types not orthogonal to M
1
, let M
1

P
M
2
means M
1
M
2
and for every p P and c M
2
,
c
M
1
p.
4) If (in (3)), P = P
M
1
(
p
a
)
we may write
_
p
a
_
instead P, similarly if P = P
M
1
(
p,r
a
)
we
may write
_
p,r
a
_
.
2.6 Claim. :
1) From tp
1
__
p
a
_
, M
we can compute tp
1
__
p
a
_
, M
if Dp(p) < .
2) From tp
2
__
p,q
a
_
, M
we can compute tp
2
__
p,q
a
_
, M
if Dp(p) < .
3) From tp
3
__
b
a
_
, M
we can compute tp
3
__
b
a
_
, M
if Dp(
b/ a) < .
4) If Denition 2.5(2) we can replace some (c,
b) r(M) by every (c,
b)
r(M).
Proof. 1),2),3) We prove this by induction on . By the denition.
4) Left to the reader.
2.7 Observation. From tp
(OB, M) we can compute tp
[OB, M] and tp
[OB, M]
is well dened if and the former is well dened.
2.8 Lemma. For every ordinal the following holds:
(1) tp
1
is well dened
5
(2) tp
2
is well dened.
(3) tp
3
is well dened.
(4) If a M
1
,
_
p
a
_

1
, M
1

(
p
a
)
M
2
then
tp
1
__
p
a
_
, M
1
= tp
1
__
p
a
_
, M
2
.
(5) If a M
1
,
_
p,r
a
_

s
2
, M
1

(
p
a
)
M
2
then
tp
2
__
p,r
a
_
, M
1
= tp
2
__
p,r
a
_
, M
2
.
(6) If a
b M
1
,
_
b
a
_

c
3
, M
1

(
b
a
)
M
2
then
tp
3
_
_
b
a
_
, M
1
_
= tp
3
_
_
b
a
_
, M
2
_
.
5
i.e. in all the cases we have tried to dene it in Denition 2.9
(
4
0
1
)

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

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

Proof. We prove it, by induction on , simultaneously (for all clauses and parame-
ters).
If is zero, they hold trivially by the denition.
If is limit, they hold trivially by the denition and induction hypothesis. So
for the rest of the proof let = + 1.
Proof of (1)
. If p has depth zero check directly.

If p has depth > 0 - by (3)
(i.e. induction hypothesis) no problem.

Proof of (2)
. Like 1.27 (and (4)
).
Proof of (3)
. Like (2)
.
Proof of (4)
. Like 1.26 (and (3)
, (6)
).
Proof of (5)
. By (2)
we can look only at (c,
b
+
) in M
1
, then use (4)
.
Proof of (6)
. By (5)
.
2.8
2.9 Lemma. For an ordinal restricting ourselves to the cases (the types p, p
1
being) of depth < :
(A1) Assume
_
p
a
_

1
, a a
1
M, a
1
is -nite,
a
1
a
p and p
1
is the station-
arization of p over a
1
.
Then from tp
1
__
p
a
_
, M
we can compute tp
1
__
p
1
a
1
_
, M
.
(A2) Under the assumption of (A1) also the inverse computations are O.K.
(A3) Assume
_
p
a
_

1
for = 1, 2, a M and p
1
p
2
.
Then from tp
1
__
p
1
a
_
, M
(and tp(( a, c
1
, c
2
), ) where c
1
, c
2
realizes p
1
, p
2
respectively, of course) we can compute tp
1
__
p
2
a
_
, M
.
(B1) Assume
_
p
,r
a
_

sc
2
for = 1, 2, a M and p
1
p
2
.
Then (from the rst order information on a, p
1
, p
2
, r
1
, r
2
, of course, and)
tp
2
_
_
p
1
,r
2
a
1
_
, M
_
we can compute tp
2
__
p
2
,r
2
a
_
, M
.
(B2) Assume a a
1
M,
a
1
a
p,
_
p,r
a
_

cs
2
, r r
1
S( a
1
), r
1
does not fork
over a,
_
so
_
p,r
1
a
1
_

2
_
.
Then from tp
2
__
p,r
1
a
_
, M
we can compute tp
2
__
p,r
2
a
_
, M
.
(
4
0
1
)

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

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

56 SAHARON SHELAH
(B3) Under the assumption of (B2), the inverse computation are O.K.
(C1) Assume
_
b
a
_

c
3
, a
b M, a a
1
,
a
a
1
,
b
1
= ac( a
1
+
b).
Then from tp
3
__
b
a
_
, M
we can compute tp
3
__
b
1
a
1
_
, M
.
(C2) Under the assumptions of (C1) the inverse computation is 0.K.
(C3) Assume
_
b
a
_

3
,
b b
,

b
b

a
a,
= ac(
). Then from tp
3
__
b
a
_
, M
we
can compute
_
tp
3
_
_
a
_
, M
_
:

b b
M and

b
b
=

b
b
_
.
Proof. We prove it, simultaneously, for all clauses and parameters, by induction on
and the order of the clauses.
For = 0: easy.
For limit: very easy.
So assume = + 1.
Proof of (A1)
. As p is stationary
a
1
a
, for every c p(M),
c
a

c
a
1
, which
necessarily is p
1
, hence p(M) = p
1
(M). Also the dependency relation on p(M) is
the same over a
1
, hence dimension. So it suces to show:
() for c p(M), from tp
3
__
ac( a+c)
ac a
_
, M
we can compute
tp
3
__
ac( a
1
+c)
ac a
1
_
, M
.
But this holds by (C1)
.
Proof of (A2)
. Similar using (C2)
.
Proof of (A3)
. If p
1
(equivalently p
2
) has depth zero the dimensions are equal.
Assume they have depth > 0 hence are trivial and dependency over a is an equiv-
alence relation on p
1
(M) p
2
(M).
Now for c
1
p
1
(M), from tp
3
__
ac(a+c
1
)
ac( a)
_
, M
we can compute for every complete

type over ac( a + c
1
) not forking over a, and

d realizing r, tp
3
__
ac( a+
d+c
1
)
ac( a+
d)
_
, M
-
by (C1)
, then we can compute for each such r,

d,
(
4
0
1
)

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

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

_
tp
3
__
ac( a +

d + c
2
)
ac( a +

d)
_
, M
_
: c
2
p
2
(M) and
c
2
ac( a +

d + c
1
)

a
( a +

d)
(necessarily c
2
d)
_
(this by (C3)
).
Proof of (B1)
. As in earlier cases we can restrict ourselves to the case Dp(p
) > 0.
We can nd (c
) r
(M),
b
1
b
2
, c
1
b
1
b
2
(by [Sh:c, X,7.3](6)]. By 2.8(2) (and
the denition) from tp
2
__
p
1
,r
1
a
_
, M
we can compute that it is equal to

tp
1
__
c
1
/ac( a+
b
1
)
ac( a+b
1
)
_
, M
.
By (A1)
we can compute tp
1
__
c
1
/ac( a+
b
1
+
b
2
)
ac( a+b
1
+
b
2
)
_
, M
hence by (A3)
we can com-
pute tp
1
__
c
2
/ac( a+
b
1
+
b
2
)
ac( a+
b
1
+
b
2
)
_
, M
.
Now use (A2)
to compute tp
1
__
c
2
/ac( a+
b
2
)
ac( a+
b
2
)
_
, M
and by 2.8(2), 2.4(2) it is equal

to tp
2
__
p,r
a
_
, M
.
Proof of (B2)
. Choose (c,
b) r(M) such that c
a
a
1
.
From tp
2
__
p,r
1
a
_
, M
we can compute tp
1
__
c/( a+
b)
a+
b
_
, M
(just see 2.8(2) and

Denition 2.4), from it we can compute tp
1
__
c/( a+
b+ a
1
)
( a+
b+ a
1
)
_
, M
(by (A1)
), from it
we can compute tp
2
__
p,r
2
a
2
_
, M
(see 2.8(2) and Denition 2.4).

Proof of (B3)
. Let
_
p,r
b
1
_

s
r
, p a
1
be given. So necessarily
a
1
a
p (this to
enable us to use (B2, 3). It suces to compute tp
2
__
p,r
b
1
_
, M
and we can discard

the case Dp(p) = 0.
So p is regular
b
1
, a
1
, hence p
b, p a, and as a
b,
b = ac(
b) we can nd
r,
_
p,r
1
b
_

2
, (see 2.3) and we know tp
2
__
p,r
1
b
_
, M
, and we can nd r
2
, a complete
type over

b
1
extending r
1
which does not fork over

b
1
. From tp
2
__
p,r
1
b
_
, M
we can
compute tp
2
__
p,r
2
b
1
_
, M
by (B2)
, and from it tp
2
__
p,r
b
1
_
, M
by (B1)
.
Proof of (C2)
. Similar, use (B3)
instead of (B2)
.
(
4
0
1
)

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

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

58 SAHARON SHELAH
Proof of (C3)
. Without loss of generality

b
b
is semi regular, let p
be a regular
type not orthogonal to it and without loss of generality Dp(p
) > 0

b
b
regular
(as in 2.3).
If p
has depth zero, then the only p appearing in the denition tp

3
__
b
a
, M
_
is p
(up to ) and this is easy. Then tp

2
is just the dimension and we have no

problem.
So assume p
has depth > 0. We can by (B1)
, (B2)
compute tp
2
__
p
,q
_
, M
when p
b, p
(regardless of the choice of

b
). Next assume p
; by (B1)
without loss of generality q
does not fork over

b. As Dp(p
) > 0, it is trivial (and

we assume w
p
(
b) = 1) hence

b
b is regular so in tp
2
__
p,q
_
, M
we just lose a
weight 1 for one specic tp
3
type: the one

b
realizes concerning which we have a

free choice. We are left with the cases p
b, p
; well we know tp
3
but we have
to add tp
3
? Use Claim 2.6(3) (and (A1)
as we add a parameter).
2.9
2.10 Claim. tp
3
__
b
a
_
, M
, tp
3
[ a, M], tp
3
[M] are expressible by formulas in

L
(d.q.).
By 2.9 we have
2.11 Conclusion. If Dp(T) < then:
1) From tp
3
__
B
A
_
, M
we can compute tp
__
B
A
_
, M
(the type from 1).

2) Similarly from tp
3
[A, M] we can compute tp
[(A), M].
From 2.6, 2.10, 2.11 and 1.30 we get
2.12 Corollary. If = Dp(T) and M, N are
-saturated, then
M

= N tp
3
[M] = tp
3
[N] M

=
L
(d.q.)
N.
(
4
0
1
)

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

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

Appendix
The following claries several issues raised by Baldwin. A consequence of
the existence of nice invariants for characterization up to isomorphism (or

characterization of the models up to isomorphism by their L-theory for
suitable logic L)
naturally give absoluteness, e.g. extending the universe say by nice forcing preserve
non-isomorphism. So negative results for
() is non-isomorphism (of models of T) preserved by forcing by nice forcing
notions?
implies that we cannot characterize models up to isomorphism by their L-theory
when the logic L is nice, i.e. when Th
L
(M) preserved by nice forcing notions.
So coding a stationary set by the isomorphism type can be interpreted as strong
evidence of no nice invariants, see [Sh 220]. Baldwin, Laskowski, Shelah [BLSh
464] show that not only for every unsuperstable; but also for some quite trivial
superstable (with NDOP, NOTOP) countable T, there are non-isomorphic models
which can be made isomorphic by some ccc (even -centered) forcing notion. This
shows that the lack of a really nite characterization is serious.
Can we still get from the characterization in this paper an absoluteness result?
Note that for preserving
-saturation (for simplicity for models of countable T)

we need to add no reals
6
, and in order not to erase distinction of dimensions we
want not to collapse cardinals, so the following questions is natural, for a rst order
(countable) complete T:
()
T
assume V
1
V
2
are transitive models of ZFC with the same cardinals and
reals, the theory T V
1
. If the models M
1
, M
2
are from V
1
and they are
models of T not isomorphic in V
1
; must they still be not isomorphic in V
2
7
()
T,
like ()
1
T
we assume in addition P()
V
1
= P()
V
2
.
2.13 Theorem. 1) For countable rst order complete T the answer to ()
T
and
()
T,
for any is negative except when possibly T is superstable, NDOP, NOTOP.
2) For any rst order complete T for the class of
-saturated models, the answer

to ()
T,|T|
is negative except possible when T is superstable with NDOP.
6
(the set of {ac( a) : a
>
M} is absolute but the set of their enumeration and of the
{f (ac( a)) : f AUT(C), f( a) = a} is not).
7
Note we did not say they have the same -sequences of ordinals, e.g. if V
2
= V
P
1
, P Prikry
forcing, then the assumption of ()
T
holds though a new -sequence of ordinals was added. So
for V
1
V
2
as in ()
T
, the L
,
1
-theory is not necessarily preserved.
(
4
0
1
)

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

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

60 SAHARON SHELAH
Proof. By quoting.
So we restrict ourselves to these. It should be quite transparent that L
,
(q.d.)-
theory is preserved from V
1
to V
2
(as well as the set of sentences in the logic).
Hence
2.14 Theorem. For the class of
-saturated models of superstable NDOP, NO-

TOP theory T the answer to ()
T,|T|
is yes.
Proof. In V
1
for = 1, 2 let A
= A M
eq
: A is -nite and acl(A) = A.

Clearly the same denition gives A
in V
2
and M
1
, M
2
are
-saturated also in V
2
.
A A
[A[ [T[ and let F
= f: for some A
1
A
2
and A
2
, A
2
, f is a one-
to-one function from A
1
onto A
2
which is an (M
1
, M
2
)-elementary mapping. Again
this denition gives the same set. We can dene a rank function rk:F
Ord
such that rk(f) = i (M
1
, a)
a Dom(f)

L
,
,(d.q.)
(M
2
, f(a))
a Dom(f)
and
it too is absolute.
Easily in both universes
(a) M
1

= M
2
i M
1

L
,
(d.q)
M
2
.
[Why? By Theorem 1.5.]
(b) M
2

L
,
0
(d.q)
M
2
i there is F F
as in
M
1
,M
2
from 1.4.
[Why? By 1.4.]
(c) there is F F
as in
M
1
,M
2
from 1.4 i belongs to the range of the
rank function.
[Why? Well known.]
(
4
0
1
)

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

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

REFERENCES.
[BLSh 464] John T. Baldwin, Michael C. Laskowski, and Saharon Shelah. Forc-
ing Isomorphism. Journal of Symbolic Logic, 58:12911301, 1993.
math.LO/9301208.
[BeSh 307] Steven Buechler and Saharon Shelah. On the existence of regular types.
Annals of Pure and Applied Logic, 45:277308, 1989.
[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 225] Saharon Shelah. On the number of strongly
-saturated models of
power . Annals of Pure and Applied Logic, 36:279287, 1987. See also
[Sh:225a].
[Sh:c] Saharon Shelah. Classication theory and the number of nonisomorphic
models, volume 92 of Studies in Logic and the Foundations of Math-
ematics. North-Holland Publishing Co., Amsterdam, xxxiv+705 pp,
1990.

Saharon Shelah - Characterizing An - Epsilon Saturated Model of Superstable NDOP Theories by Its L, - Epsilon Theory

Uploaded by

Document Information

Original Description:

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 - Characterizing An - Epsilon Saturated Model of Superstable NDOP Theories by Its L, - Epsilon Theory

Uploaded by

Copyright:

Available Formats

(

-saturated model of the theory is

(d.q.) (essentially, in C we can quantify over sets which are included in

-saturated model M of a superstable DNOP theory

for and (the last is not essential but claries)

: I) is non-forking enough: for every I the set a

are chosen then to a large extent M

). So we consider a pair (A, B) where

) and we try to dene the

(b) it expresses the essential information in M

being -nite, so in stage n + 1, choosing a

we cannot take care of all types

so addition theorem takes care. So though we are thinking on

-saturated), we get just a decomposition.

-saturated models of an NDOP su-

-saturated models rather than all models. The rst example is

-saturated models of shallow (superstable NDOP) theories T,

for all . Proposition 1.14 contains a number of

-saturated structures that have

are isomorphic. The induction step is carried out in 1.35.

= ; for I a set of sequences ordinals we let

-atomic over A if for every nite sequence

b, A), equivalently for some -nite

is slightly stronger than L

but we have cardinality quantiers and moreover di-

consist of the formulas of L

such that has quantier depth

but we have cardinality quantiers and moreover di-

b B is such that tp( a, B) does not fork over

b, then trivially ac(

b) forks over B, hence

instead of F and may use implicitly the consequences

there is a non-empty family F such that:

( = 1, 2) then for some g F we have:

-saturated model characterizes it up to iso-

-saturated models of T (so M

: I) of N, in the sense that for some

-prime over M + a and tp(a, M) is regular. But this is too large

B and B/M(= B/M

) does not fork over A. We

-prime over M + a) we have stp

; we can think of this as advancing in

) for some an immediate

) is regular and is orthogonal to A (as in the decomposition we require

is the transitive closure of

: 1 < n) for some n 1, satisfying

there are k < and

(which necessarily has domain

, check) with domain B

: k + 1) we repeat the argument (of

: n)) in the proof of 1.14(6) above, i.e., choose B

[A, M], for A an -nite subset of M:

[A, M]) where

-saturated model, and we choose M

-prime over and let I

be a maximal subset of c M : tp(c, M

-prime models (and the local character

: I, its rst order type is

: ( ) and I are independent over N

: I, now if M = N we are done decomposing M, if not

. Choose of minimal length hence

. By properties of regular types, without loss of generality tp(c, N)