Saharon Shelah - Dependent Theories and The Generic Pair Conjecture

(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

DEPENDENT THEORIES AND THE GENERIC PAIR
CONJECTURE
SH900
SAHARON SHELAH
Abstract. We try to understand complete types over a somewhat saturated
model of a complete rst order theory which is dependent, by decomposition
theorems for such types. Our thesis is that the picture of dependent theory
is the combination of the one for stable theories and the one for the theory
of dense linear order or trees (and rst we should try to understand the quite
saturated case). As a measure of our progress, we give several applications
considering some test questions; in particular we try to prove the generic pair
conjecture and do it for measurable cardinals.
Annotated Content
0 Introduction, pg.3-8
1 Non-splitting construction, pg.9-17
[For -saturated M and N such that M N we try to analyze N over
M by nding M
1
, N
1
such that M M
1
N
1
, M N N
1
and both
M
1
/M, N
1
/N are understood but in opposite ways. The rst similar in
some sense to the stable situation, the second to the situation for order.]
2 The type decomposition theorem, pg.18-34
[For -saturated M C and

d C of length <
+
we try to analyze the type
tp(
d, M) in two steps - pseudo stable and tree-like one. This is the main
aim of the section (and a major part of the paper). It is done by looking
at K
and mxK
,,
. A consequence which fullled to some extent the
aim is the Type Decomposition Theorem (2.4). As a second consequence
we give a characterization of M is exactly -saturated, > cf() >
[T[, see 2.2. In fact, we deal a little with singular exact saturation per
se. Unfortunately there are independent (complete rst order theories)
T which has no model with singular exact saturation, see 2.22. But the
existence of an indiscernible set for dependent T suce (see 2.25 under
instances of GCH) and has a neat characterization. Also, if p is a complete
1-type over a model M of T which is quite saturated then p has a spectrum
in a suitable sense, see 2.30.]
Date: April 7, 2011.
The author would like to thank the Israel Science Foundation for partial support of this research
(Grant No. 242/03). I would like to thank Alice Leonhardt for the beautiful typing. First version
as F756 was 2006/Feb/14; and 900 - 2007/Jan/10.
1
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

2 SAHARON SHELAH
3 Existence of strict decomposition, pg.35-41
[E.g. here a saturated model M of cardinality , a measurable cardinal, is
analyzed. What we get is a better decomposition theorem (the strict one).]
4 Consequences of strict decomposition, pg.42-53
[We start by sucient conditions for a sequence being indiscernible. For a
measurable cardinal (> [T[) we conrm the structure half of the generic
pair conjecture. Toward this if we have the consequences of 3 we can
analyze generic pairs of models of T in . In a slightly simplied formulation
this means: if 2
=
+
, ( =
<
> [T[), M
, a model of T of cardinality
for <
+
is -increasing continuous, M = M
: <
+
is
+
-
saturated, then for a club E of
+
for all < belonging to E :
has conality the pair (M
, M
) has the same isomorphism type. In

fact for
1
<
2
we get L
,1
(
T
)-equivalence, so we have a derived rst
order theory. For our proof we show that an increasing (short) sequence of
so called strict (, )-decompositions has a limit.]
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

DEPENDENT THEORIES AND THE GENERIC PAIR CONJECTURE SH900 3
0. Introduction
We rst give a page of introductory remarks for non-logicians to clarify notions
and to motivate working on dependent theories. The classical center of model
theory is investigating elementary classes, i.e. we x a vocabulary (i.e. a set
of predicates and function symbols), for a -structure M let Th(M) be the set of
rst order sentences which M satises, a complete rst order theory T is Th(M)
for some -model M. We x T and =
T
and investigate -models of T, i.e.
-structure M such that T = Th(M); about other contexts, see e.g. [Sh:E53].
Let M, N denote such structures and they are called models. Let a,
b, c,

d denote
sequences of elements of such models and ( x) or ( x, y) denote members of L(),
i.e. the set of rst order formulas in this vocabulary but we allow x to be innite
though the formula is nite so only nitely many x
, y
j
are relevant.
Let M [= [ a] say that the model M satises the formulas ( x) under the
substitution x a (so a, x have the same length).
The right notion of sub-models is , being elementary submodel where M N
i M N and for every ( x) L() and a
g( x)
M we have M [= [ a] i
N [= [
b].
Recall that an ordinal is the isomorphism type of a well ordering (which is a
linear order for which every non-empty set has a rst member). But we identify
an ordinal with the set of smaller ordinals. Also a cardinal is an ordinal with no
smaller ordinal of the same power. Here a cardinal means innite one if not said
otherwise. Let
be the -th innite cardinal and the cardinality [U [ of a set U

is the minimal ordinal of the same power.
Now for transparency assume G.C.H., the generalized continuum hypothesis, i.e.
2
=
+1
for every ordinal , and the successor
+
of a cardinal is
+1
when
=
.
We say E is a closed subset of the limit cardinal when E and < =
sup( E) E and E is called unbounded when ( < )()( E),
E is a club of is the shorthand for E is a closed unbounded subset of .
For an ordinal let cf() = min[C[ : C an unbounded subset of =
minotp(C) : C a closed unbounded subset of ; we say is regular if = cf() is
innite (hence is a cardinal), now recall that if is a limit ordinal (e.g. a cardinal)
then cf() is regular, and every cardinal of the form
+
is regular.
A central notion is type; for A M and a a sequence from M let tp( a, A, M)
be the set ( x,
b) : ( x, y) L(),
b a sequence from A and M [= [ a,
b].
Let
S
(A, M) = tp( a, A, N) : M N, a a sequence of length from N

S
(M) = S
(M, M).
By this we can dene another central notion. M is -saturated i for every A
M, [A[ < and p S
1
(A, M) some a M realizes p in M which means p =
tp(a, A, M). We say the model M is saturated when it is -saturated of cardinality
. Let EC
(T) be the class of models of T of cardinality .

It is classically known that for [T[, (as we are assuming 2
=
+
) there is a
saturated member of EC
+(T), it is unique up to isomorphism, and the union of an

(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

4 SAHARON SHELAH
-increasing chain of saturated members of EC
+(T) of length
+
is a saturated
member of EC
+(T). On the background so far, see e.g. Chang-Keisler [CK73].

A major theme of the authors work is trying to nd natural dividing lines (i.e.
properties) in the family of rst order complete T, in particular inside denitions
by formulas and outside denition by properties of the class of its models. Such a
property is interesting as a divding line when we have consequences for those with
the property and for those without it; see e.g. [Sh:E53, (1A)].
A major such dividing line is T is stable which holds i ()
1
T
i ()
2
T
where
()
1
T
for some ( x, y) L(
T
), model M of T and a
n

(g( x))
M,
b
n

(g( y))
M
for n < we have n < m M [= [ a
n
,
b
m
]
()
2
T
for every [T[ and limit ordinal of conality > [T[, the union of any
-increasing chain of saturated models of T of cardinality is saturated.
Another major dividing line is T is superstable which holds i
()
3
T
like ()
2
T
allowing any limit ordinal .
On this and the relevant history, see e.g. [Sh:c].
The property we deal with is T is dependent, also called T is NIP, where
its negation, T is independent or T has the independence property means
()
4
T
there are ( x, y) L(
T
), a model M of T and a
u

(g( x))
M,
b
n

(g( y))
M
for u , n < such that n u M [= [ a
u
,
b
n
].
What is the motivation to investigate this dividing line? First, it has a nice, simple
denition, parallel to the one for stable theories. Second, it is a wider class than
that of the stable theories; also, extremely important for many, whereas elds with
stable rst order complete theory are hard to come by (algebraically closed and
separably closed are the only known ones), there are many important elds with
dependent rst order complete theory (the p-adics and many of the power series
elds). Third, there are some results on it indicating it is not unreasonable to hope
there is a rich theory on it to be discovered.
On background on dependent theories, see [Sh:715], [Sh:783].

Let T be a xed rst order complete theory. For transparency we assume G.C.H.,
i.e. 2
=
+
for every innite cardinal and consider only regular > [T[. Let
M
: <
+
) be an -increasing continuous sequence of models of T of cardinality
with M being saturated where M := M
: <
+
being saturated. Now M is
unique (up to isomorphism, for each ) and though

M is not unique, for any two
such sequences

M
,

M
there is a closed unbounded subset E of

+
and isomorphism
f from M
= M
: <
+
onto M
= M
: < such that f maps M
onto M
for every E.
So it is natural to ask ( > [T[ regular and E varies on closed unbounded subsets
of
+
)
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

1
what
1
is n
(T) := Min
E
[M
/

=: E[? where M
/

= is the isomor-
phism type of M
. When is n
(T) equal to one?

Now (see [Sh:868]):
2
n
(T) = 1 i T is superstable
3
for countable T, n
(T) = 2 i T is strictly stable (i.e. T is stable, not

superstable)
4
given an ordinal , for large enough n
(T) = [ + 1[ if T is stable and

(T) =
(recalling that for a stable T, (T) is cardinal [T[

+
, so for
countable T it is
0
or
1
)
5
if T is unstable, =
then n
(T) [ + 1[.
[Why? Because for some closed unbounded subset E of
+
, if E then M
is
cf()-saturated but not (cf()))
+
-saturated hence [
1
,
2
E cf(
1
) ,= cf(
2
)
M
1
M
2
].]
Hence it is natural to replace n
(T) by
6
let n
,
(T) = Min
E
[M
/

=: E and cf() = [ when > = cf()
(and > [T[ regular, and as above E varies on the clubs of
+
).
Below we use n
,
(T) only when > [T[ + = cf() and remember that for
simplicity we are assuming G.C.H.
Now (see [Sh:868]):
7
if T is stable then n
,
(T) = 1.
It is natural to ask whether this characterizes stable theory. The answer is no, in
fact, by an example everyone knows (by [Sh:877, 1])
8
n
,
(T) = 1 for T = Th(Q, <), the theory of dense linear orders with
neither rst nor last element, when is regular so =
<
> = cf()
During the proof we analyze p S(M
), M
saturated. So M
is a linear order
and p induces a cut (C
p
, C
+
p
) of M
, i.e. C
p
= a M
: (a < x) p is an initial
segment of M
and its compliment, a M
: (a < x) / p is an end segment.

This gives a pair of conalities, (
p
,
+
p
),
p
the conality of the linear order C
p
and
+
p
the conality of the inverse of C
+
p
.
Now
()
8.1
if
p
:= min
p
,
+
p
< , then the type is determined by any subset of
M
of cardinality
p
such that
the set unbounded in C
p
if
p
=
p
and
unbounded from below in C
+
p
if
p
=
+
p
()
8.2
if
p
= , and we expand M
by the (unary) relation C

1
p
, we still get a
saturated model.
1
We can present the problem dierently, about the existence of (variations of) (, )-limit
models (so 2
=
+
is no longer necessary, by forcing this is equivalent). Also, instead of the
function n getting the value
+
we can consider saying for some club no two relevant cases
are isomorphic. This does not make a real dierence but we nd the present choice has more
transparent presentation.
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

6 SAHARON SHELAH
Next considering
7
+
8
you may think that for every T we get n
,
(T) = 1, but
([Sh:877, 2.3(2)] implies directly that):
9
n
,
(T) =
+
if T is Peano arithmetic
moreover, this holds for quite many theories T (by [Sh:877, 2]):
10
n
,
(T) =
+
if T has the strong independence property (see [Sh:72], i.e.
for some rst order formula (x, y), (M, a) : a M) is an independent
sequence of subsets of M).
For me this rings a bell and strengthens a suspicion - maybe the dividing line is T
independent/T dependent, indeed (by [Sh:877, 2]):
11
n
,
(T) =
+
if T is independent, a successor cardinal.
We try to address the complement, the structure side. This calls for analyzing
appropriate -increasing continuous sequence

M = M
i
: i ) of models of T
of cardinality . Clearly in the relevant cases they increase fast enough and M
i
is
saturated for i non-limit. Now among such sequences, is it not reasonable to rst
deal with the case of length 2?
This leads to the generic pair conjecture which says that for =
<
> [T[, we
have T is independent i n
,2
(T) =
+
where:
12
n
,2
(T) := Min
E
[(M
, M
)/

=: < belongs to E and cf() = =
cf()[.
Note that in dening n
,
(T), Reg [
0
, ] we speak on models of T whereas
here we deal with pairs of model. But for Reg [
0
, ), we naturally are
drawn to deal with -tuples, this seems close enough. Anyhow the numbers depend
just on T (and , or 2), because up to isomorphism there is one saturated model
of T of cardinality
+
.
This connects us to the long term goal of classifying rst order theories by
good dividing lines, ones in which we nd outside properties (like here investi-
gating n
,
(T) or just n
,
(T), trying to characterize it) with inside denitions
(like being dependent), and developing an inside theory; here - looking at decom-
position (in 1 decompositions of models, in 2 decomposition of types, in 3,4
strict decomposition of types). More fully for this we have to analyze types. In
1 we make a rst attempt; more exactly see 1. We try to analyze N := M
over
M := M
by trying to nd models M
1
, , N
1
such that:
1
M
= M M
1
N
1
and M
= M N = M
N
1
2
for every a
>
(M
1
) for some B
[M]
<
the type tp( a, M
, M
1
) is
denable over B
in a weak sense, i.e. does not split over B
, this means
that if n < and

b, c
n
(M
) realizes the same type over B
then so
does a
b, a c (this is parallel to ()
8.1
from
8
); it follows that similarly for
a
>
(M
1
)
3
tp(N
1
, M
1
, N
1
) is weakly orthogonal to every q S
<
(M
1
) which does not
split over some B [M
1
]
<
; the weakly orthogonal means that q has a
unique extension in S
n
(N
1
) wherever q S
n
(M
1
).
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

In 2 we try to analyze a type rather than a pair of models, also we nd it better
to deal with -types, [T[, as in the analysis we add more variables. So for
a -saturated model M C and sequence

d of length <
+
we try to analyze
tp(
d, M, C) in two steps. The rst is to add c of length <

+
such that
4
tp( c, M, C) does not split over some B M C of cardinality < .
This corresponds to the stable type (unfortunately but unavoidably depending
on ), so for the theory of dense linear orders it corresponds to types p S(M)
with
p
< , see ()
8.1
above. True, they are not really denable, but non-splitting
is a weak form of being denable. The second step is
5
tp(
d, M + c, C) is tree like, i.e. if A M C and [A[ < then for some

e

+
>
M we have tp(
d, c + e) tp(
d, A + c).
This property holds for T = Th(Q, <), p S(M) when
p
!, i.e. when both
conalities are . This is the Type Decomposition Theorem (2.4).
A consequence is some clarication of models of M of a dependent theory which
are exactly -saturated for singular . We deal with this question to some extent
per se.
In 3 we get a better decomposition - strict decomposition. But at present with
a price, assuming e.g. = |M| is a measurable cardinal. The main point appears
in 4, the existence of limits of increasing sequences of strict decompositions.
From this we get the pair genericity conjecture, the structure side for the case of a
measurable cardinal. The measurability assumption seems undesirable. Describing
this to Udi Hrushovski he was more concerned about also having the non-structure
side for independent T. Now at the time in [Sh:877] it was remarked that a similar
proof should work for the strongly inaccessibles, but the author was not motivated
enough to really look into it. Subsequently [Sh:906] completes it.
The order of the sections is by their conceptions, so there are some repetitions.
In [Sh:950] and Kaplan-Shelah [KpSh:946] we continue this work.
We thank the referee and John Baldwin for much helpful criticism.
{0n.1}
Context 0.1. 1) T is complete rst order theory.
2) C = C
T
is a monster for T, omitting T when no confusion arises; i.e. is a
large enough cardinal, C is a -saturated model such that we deal only with models
M C, sets A C of cardinality < and sequences a,
b, c,

d, e from

C for some
< . So tp( c, A) means tp( c, A, C).
3) We may not pedantically distinguish a model M and its universe, the cardinality
|M| of M is that of its universe.
{0n.4}
Notation 0.2. 1) For M C and a

M let M
[ a]
be the expansion of M by every
R
( x, a)
= (M, a) where (M, a) :=
b
g( x)
M : C [= [
b, a] for ( x, y) L(
T
)
such that g( y) = , g( x) < or pedantically ( x, y u) for x, y as above, u
nite. We dene M
[A]
similarly, i.e. as the expansion of M by R
( x, a)
= (M, a)
for every
g( y)
A and ( x, y) L(
T
).
2) Writing ( x, y) L(
T
), here is always rst order but x and y may be innite,
though sometimes are nite (said or clear from the context). Let p( x), q( x), r( x)
denote types over some A C, i.e. set of formulas of the form ( x,
b),
b
(g(
b))
A.
3) EC
(T) is the class of models M of T (so M C) of cardinality .

4) A + c is A Rang( c), etc.
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

8 SAHARON SHELAH
{0n.8}
Denition 0.3. 1) If a
t

C for t I and D is a lter on I and x = x
i
: i < )
and A C then Av( a
t
: t I)/D, A) = ( x,
b) :

b
>
A and the set t I :
C [= [ a
t
,
b] belongs to D. Note that if T is dependent, I is a linear order with

no last members and a
t
: t I) is an indiscernible sequence, see below then the
result S
(A). Also note that if D is an ultralter on I then Av( a

t
: t I)/D, A)
belongs to S
(A).
1A) Recall that if D is a lter on a
t
: t I

A we dene Av(D, A) similarly
and if I a linear order and D is the lter of co-bounded subsets of I we may omit
it.
2) If p( x), q( y) are complete types over A we say p( x), q( y) are weakly orthogonal
when for every a
1
, a
2
realizing p( x) and

b
1
,
b
2
realizing q( y) we have tp( a
1
b
1
, A) =
tp( a
2
b
2
, A).
3) For a linear order I, a
s
: s I) is an indiscernible sequence over B when g( a
s
)
is constant and if s
0
<
I
. . . <
I
s
n1
an t
0
<
I
. . . <
I
t
n1
then the sequences
a
s0
. . . a
sn1
and a
t0
. . . a
tn1
realize the same type over B.
Recall also (see [Sh:c, Ch.II,4])
{0n.17}
Fact 0.4. If T is dependent then for any formula = ( x, y, z) L(
T
) there is
n = n
< (depending on T) such that:

(a) for no c
g( z)
C and

b
i

g( y)
C for i < n, is the sequence ( x,
b
i
, c) :
i < n) independent, i.e. every non-trivial Boolean combination of the sets
(M,
b
i
, c) = a
g( x)
M : M [= [ a,
b
i
, c] for i < n is non-empty.
(b) If
b
i
: i < n) is an indiscernible sequence over C, g(
b
i
) = g( y), c
g( z)
C
(all in C) then for no a
g( a)
M do we have C [= [ a,
b
i
, c]
if( even)
for < n.
(c) Also there is a nite
L(
T
) such that in clause (b) it is enough to
demand that
b
i
: i < n) is a -indiscernible sequence.
Lastly, we quote Erd os-Rado [ER69].
{0n.19}
Fact 0.5. The -System Lemma for nite sets.
For every natural numbers k, n there is a natural number m such that: if u
i
is a
nite set with k elements for i < m then there are sets w 0, . . . , m1 with
[w[ = n and u
such that u
i
: i w) is a -system with heart u
, which means
that i ,= j w u
i
u
j
= u
.
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

1. Non-splitting constructions
On such constructions including F
nsp
see [Sh:c, Ch.IV,1,3] but F

nsp
here is F
p
there; and see [Sh:715, 4.23-4.26], however this section is self-contained.
We try here to analyze -saturated M N, e.g. by nding M
1
, N
1
such
that M M
1
N
1
, N N
1
where M
1
is F
nsp
-constructible over M, see be-

low and tp(N
1
, M
1
) is weakly orthogonal to any type over M
1
realized in some
F
nsp
-construction over it, see Theorem 1.9, part (B) noting that M, N, N
1
, M
1
here
stands for A, A
+
, M, N there. We rst recall the denition of non-splitting and
some properties.
{3k.0.4}
Denition 1.1. We say p( x) does not split over A when : if ( x,
b), ( x, c) p( x)
then tp(
b, A) ,= tp( c, A).
{3k.0.7}
Fact 1.2. 1) If
A,B,C
below holds and p( x) S
m
(B) does not split over A, then
there is one and only one q( x) S
m
(C) extending p( x) and not splitting over A
(also called the non-splitting extension of p( x) over C), where:
A,B,C
(a) A B C
(b) for every c
>
C there is

b
g( c)
B realizing tp( c, A).
2) Let I be a linear order. If tp( a
t
, B
a
s
: s <
I
t) does not split over B and
increases with t I then a
t
: t I) is an indiscernible sequence over B.
3) If tp( a, B) does not split over A, the sequence
b
t
: t I) is an indiscernible
sequence over A (or just over some A
A) and

b
t
B for t I then
b
t
: t I)
is an indiscernible sequence over A a.
4) If A B then the number of p S
(B) which does not split over A is

2
2
|A|+|T|+
, moreover if T is dependent the number is 2
|A|+|T|+
.
Proof. 1) By [Sh:3] or see [Sh:715] or see [Sh:300a, 1.10] for uniqueness.
2) By [Sh:c, I] or [Sh:300, I] or [Sh:300a, 3.2].
3) By the denitions.
4) The rst conclusion is easy and see [Sh:3] or [Sh:300a, 1], the second holds by
[Sh:783, 5.26].
1.2
{3k.2}
Fact 1.3. [Assume T is dependent.]
If p( x) is an -type over B A then we can nd q( x) S
(A) extending p( x)
such that for some C A of cardinality [T[ the type q( x) does not split over
B C.
Proof. [Sh:c, III,7.5,pg.140] or see [Sh:715, 4.24].
1.3
{3k.4}
Observation 1.4. For regular.
1) If A B, [A[ < and a
>
C and tp( a, B) is nitely satisable in A then it
does not split over A.
2) If A B, c
>
C and tp( c, B) does not split over A and i < g( c) then
tp(c
i
, B c
j
: j < i) does not split over Ac
j
: j < i. Similarly for c
j
: j i)
when j i Rang( c
j
) Rang( c).
3) If tp( c
k
, B + c
0
+ . . . + c
k1
) does not split over A B for k < n then
tp( c
0
. . . c
n1
, B) does not split over A.
4) If A A
1
B
1
B and Rang( c
1
) Rang( c) and tp( c, B) does not split over
A then tp( c
1
, B
1
) does not split over A
1
.
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

10 SAHARON SHELAH
Proof. Easy.
1.4
As in [Sh:c, IV].
{3k.6}
Denition 1.5. 1) A is an F
nsp
-construction when :
(a) A = (A, a,

B,

A, ) = (A
A
, a
A
,

B
A
,

A
A
,
A
)
(b) a = a
: < ) = a
A
: < )
(c)

B = B
: < ) = B
A
: < )
(d)

A = A
: ) = A
A
: )
(e) A
= A a
: <
(f) B
and [B
[ <
(g) tp(a
, A
) does not split over B
.
2) We let g(A) =
A
and writing A we may omit

A
A
,
A
as they are determined
by the others so may write A = (A, a,

B) or A = (A, (a
, B
) : < )). We may

replace a
by a nite sequence a
with no real change.

3) We say the F
nsp
-construction A is -full when cf(g(A)) and if q

S(A
A
g(A)
) does not split over B where B A
A
g(A)
has cardinality < , then
< g(A) : a
realizes p A
A
and B A
A
is unbounded in
A
and has order
type divisible by .
4) We say C is F
nsp
-constructible over A when there is an F

nsp
-construction A
such that A = A
A
= A
A
0
and C = A
A
g(A)
.
{3m.1}
Denition 1.6. 1) Let A
C mean that C is F
nsp
-constructible over A.
2) We say that (A
+
, A) is -reduced when : if A
C and c
>
(A
+
) then tp( c, A)
has a unique extension to a complete type over C.
3) We say the (N, M) is -nice when :
(a) (N, M) is -reduced and M N
(b) M is -saturated
(c) N is -saturated
(d) if M
M
+
then M
[N]
M
+
[N]
, see below.
3A) Recall M
[B]
is M expanded by R
( x, a)
=
b
g( x)
M : C [= [
b, a] for
( x, y) L(
T
), a
g( y)
N and recall Th(M
[B]
) is dependent by [Sh:783, 1].
4) We say that (M, A) is pseudo -reduced when: if c
>
A, |M
1
| < , M
1

M, q( x) S
<
(M) is nitely satisable in M
1
then q( x), tp( c, M) are weakly
orthogonal.
{3m.2}
Observation 1.7. For regular:
1)
is a partial order.
2) If A
i
: i ) is increasing continuous and i < A
i

A
i+1
then A
0

.
3) In Denition 1.6(2) it is enough to consider c
>
(CA).
4) If A
B and c
>
B then
2
tp( c, A) does not split over some A
A of
cardinality < .
5) If the pair (M, C) is -reduced then (M, C) is pseudo -reduced.
6) If tp( a, A) does not split over B and B A has cardinality < then A
A+ a.
2
we could have chosen this as the denition. This changes the places we really need regular.
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

Proof. Easy (e.g. part (6) by 1.4(2) and part (4) by 1.3(3),(4)).
1.7
{3m.3}
Claim 1.8. [T is dependent and = cf() > [T[].
1) For every A there is a -saturated C such that A
C and [C[ ([A[ +[T[)

<
.
2) Similarly -full -saturated, for = cf() (clearer if [C[ ([A[ +[T[)
<
+
2
2
).
Proof. 1) By 1.3 + 1.7(2) and 1.7(6).
2) Similarly (by 1.2(4)).
1.8
Now we arrive to the rst result giving a decomposition. The type tp(A
+
, A)
is decomposed in Theorem 1.9 by nding M such that A
M, (so the complete

types over A realized in M are somewhat denable) and (A
+
, M) is -reduced, so
the type tp(A
+
, M) is weakly orthogonal to types in S
<
(M) not splitting over
subsets of M of cardinality < .
{3m.4}
Theorem 1.9. The Density of Reduced Pairs Theorem [T dependent].
For any A A
+
and = cf() > [T[ and satisfying
3
< =
[A
+
[
(A) we can nd M such that M is a model of cardinality such that A
M
and (A
+
, M) is -reduced
(B) for some M as in clause (a) and N the pair (N, M) is -reduced and even
-nice and A
+
N.
Proof. Proof of (A):
Our intention is to try to do a construction as described in below. Having
carried the induction the proof is divided to two cases. In the rst we get the
desired conclusion. In the second, we get a contradiction to T being independent;
formally to the maximality of the k chosen in (g)() of .
We choose M
i
, B
i
, j
i
, c
i
by induction on i
+
such that
(a) M
i
is -increasing continuous, M
i
of cardinality +[i[
(b) j
i
i, B
i
M
i
, |B
i
| <
(c) M
0
is F
nsp
-constructible over A
(d) M
i+1
is F
nsp
-constructible over M
i
and M
i+1
is -saturated
(e) c
i

>
(M
i+1
) and B
ji
M
i
has cardinality <
(f) tp( c
i
, M
i
i
(g) if cf(i) and subclause () below holds then subclause () holds
where
Subclause (): There are j < i, m < , B M
j
of cardinality < and
p( x) S
m
(M
i
) which does not split over B and p( x) has 2 extensions in
S
m
(M
i
A
+
).
Subclause (): There are m
i
< , j = j
i
< i, k = k
i
< and ( x, y) =
i
( x
i
, y
i
) L(
T
) with x, y nite and

b =

b
i

(g( y))
(M
j
A
+
) and
0
<
. . . <
k1
from the interval [j
i
, i) such that
tp( c
i
, M
i
)
i
( x,
b
i
) and tp( c
i
, M
i
)
i
( x,
) are consistent
tp( c
, M
) = tp( c
i
, M
) and
3
no real harm if we replace < =
|A
+
| by < =
+ 2
2
+|T|
|A
+
|
and then we can use only the rst version of 1.2(4).
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

12 SAHARON SHELAH
C [= [ c
b]
if( even)
and
k is maximal for the given ( x, y),
b, j
i
(see below; k is well dened
as T is dependent, see below)
C [= [ c
i
,
b]
if(k is even)
.
So in stage i we rst choose M
i
: if i = 0 by clause (c) such model M
i
exists by
1.8(1), if i is a limit ordinal we choose M
i
as M
j
: j < i and if i = j +1 (so c
j
has
already been dened) then choose M
i
such that M
j
c
j

M
j+1
and M
i
= M
j+1
is -saturated of cardinality (and -full if you like), possible by Claim 1.8(1).
Note that M
j

M
j
c
j
by clause (f) and 1.7(6) hence M
j

M
j+1
recalling
1.7(2).
Note
there is n = n
( x, y)
depending on ( x, y) and T only such that in subclause
() we have
i
( x
i
, y
i
) = ( x, y) k
i
n.
[Why? As in subclause () by 1.2(2) the sequence c
: < k) is an indiscernible
sequence, so by T being dependent we are done by 0.4.]
Second, why can we choose (m
i
, j
i
, B
i
,
i
, tp( c
i
, M
i
)) as required in clause (g)? If
cf(i) < or the antecedent of clause (g), i.e. (g)() fails then trivially yes (choose
e.g. c
i
as the empty sequence). Otherwise let j < i, B M
j
be of cardinality
< , m < and p( x) S
m
(M
i
) which does not split over B and which has
extensions p
0
( x) ,= p
1
( x) in S
m
(M
i
C) with p
0
(M
i
A
+
) ,= p
1
(M
i
A
+
) but
p
0
M
i
= p = p
1
M
i
does not split over B.
Hence for some

b
>
(M
i
A
+
) and = ( x, y) L(
T
) we have ( x,
b)
p
1
( x), ( x,
b) p
0
( x); as i is a limit ordinal without loss of generality

b
>
(M
j
A
+
). We now try to choose
by induction on n
( x, y)
such that:
(a) j
< i and k <

k
<
(b) c
realizes p( x) M
(c) C [= [ c
b]
if( is even)
(d)
is minimal under (a)+(b)+(c).

So by for some k n
( x, y)
we have:
is well dened i < k. At last we

choose:
() (a) B
i
= B
(b)
i
=
(c) j
i
= j
(d) k
i
= k
(e) c
i
realizes p
1
( x) if k is even and realizes p
0
( x) if k is odd.
So at last we are in a situation where the construction from (a) (g) having been
carried out. So now comes the division to cases.
Let S
= i <
+
: cf(i) and subclause () of clause (g) holds for i.
Case 1: S
is a stationary subset of
+
.
Hence for i S
, there are
i
( x
i
, y
i
),
b
i
and k
i
< and
0
(i) < . . . <
ki1
(i) <
i as in subclause () of (g) and by Fodors lemma for some m
< , j <
+
, B, ( x,
b), k
,
i
: i < k
) and a stationary subset S of S
= <
+
:
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

cf() we have S j
= j B
= B g( c
) = m
( x,
) =
( x,
b) k
= k
<k
() =
. Also without loss of generality by 1.2(4) we

have S tp( c
, M
min(S)
) = tp( c
min(S)
, M
min(S)
) recalling that the number
of such types is 2
|B|+|T|
. Choose (0) < (1) from S so both has conality
and B
(0)
= B
(1)
, tp( c
(0)
, M
(0)
) tp( c
(1)
, M
(1)
) by 1.2(1) and

b
(1)
=

b
(0)
and
i
((0)) =
i
((1)) for i < k
. But we could have chosen in stage (1),

k
for
k < k
and k
(1)
= k
+ 1 and
k
((1)) = (0), contradiction to the maximality of
k in subclause () of (g).
Case 2: Not Case 1.
Then for a club of i <
+
if cf(i) then subclause () of clause (g) fails for i
hence M
i
exemplies that we have gotten the desired conclusion in (A) of 1.9.
Proof of (B): By induction on i <
+
we choose M
i
, M
+
i
, B
i
, j
i
, c
i
such that
clauses (a),(c),(d) of and

(h) M
+
j
: j i) is -increasing continuous and A
+
M
+
0
(i) M
i
M
+
i
and M
+
i
has cardinality and if i is non-limit then
M
i
, M
+
i
are -saturated
(j) if cf(i) and there are j
i
< i, m < , B M
ji
of cardinality
< and p S
m
(M
i
) which does not split over B and has
2 extensions in S
m
(M
+
j
) then subclause () of clause (g)
above holds (with

b
g( y)
(M
+
j
)).
The rest of the proof is similar to that of (A). Alternatively, we choose M
i
, M
+
i
by
induction on i such that clauses (a),(c),(d) of and (h),(i) of
and
(k) if i = j + 1 then (M
+
j
, M
i+1
) is -reduced.
Then (M
+
, M
) are as required on (N, M).

1.9

For the rest of the section we shall assume (as we use it all the time).
{3n.0}
Hypothesis 1.10. T is dependent.
{3n.1}
Denition 1.11. 1) S
nsp
<
(A) = p S(A) : p does not split over some B A of
cardinality < .
2) S
nsp,
<
(A) S
(A) and S
nsp,<
<
(A)
<
S
(A) are dened similarly.

3) S
nsp
(A) = p S(A) : p is weakly orthogonal to r for every r S

nsp
<
(A).
4) S
nsp,
(A), S
nsp,<
(A) are dened similarly.

We may note
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

14 SAHARON SHELAH
{3n.1d}
Observation 1.12. 1) If tp( c
1
, A) belongs to S
nsp,
(A) and c
2

C and Rang( c
2
)
Rang( c
1
) then tp( c
2
, A) belongs to S
nsp,
.
2) tp( c, A) S
nsp,
(A) i tp( cu, A) S

nsp,m
(A) for every nite u g( c).

3) If tp( a, A) S
m
(A) is weakly orthogonal to tp( c, A) and does not split over
B A and every q S
<
(B) is realized in M then tp( a, A+ c) does not split over
B.
Proof. Straight.
1.12
{3n.2}
Observation 1.13. If = cf() > [T[, the model M is -saturated and p
S
m
(M), then we can nd N, q such that:
1
N
(a) |N| = |M|
<
(b) q S
m
(N) extends p
(c) N is F
nsp
<
-constructible over M
2
N,q,
(a) N is -saturated and q S
m
(N)
(b) if r S
nsp,<
<
(N) then r, q are weakly orthogonal, i.e. q S
nsp,m
(N).
Proof. Let c realize p( x) and let C = Rang( c), now we apply clause (A) of Theorem
1.9.
1.13
{3n.3}
Theorem 1.14. The Tree-like Type Theorem Assume q( x) S
nsp,
(N) and N is
-saturated and > = [T[ + [[ and let z = z
: < ). Then we can nd

a sequence

=
( x, y)
( x, z) : ( x, y) L(
T
)) of formulas such that for every
A M of cardinality < there is c

M such that:
(a)
( x, y)
( x, c) : ( x, y) L(
T
) qRang( c) q
(b) for each ( x, y) L(
T
) we have
( x, y)
( x, c) ( x,
b) :

b
g( y)
A and
( x,
b) q.
Proof. This follows from Claims 1.15, 1.16 below.
1.14
{3n.4}
Claim 1.15. 1) Assume that
2
N,q,
from the Claim 1.13 holds, which means N is
-saturated and q S
nsp
(N). Then
3
N,q,
if M N has cardinality < and (x, y) is a formula with parameters
from N, then for some (x,

d) =
(x,y),M
(x,

d
(x,y),M
) q and
M
2
we have (x,

d) p
M,
(x,y)
where
p
M,
(x,y)
= (x, b)
(b)
: b M; so it is included in q.
2) Part (1) works also for q S
m
(N), i.e. q S
nsp,m
(N) and = ( x, y) where

g( x) = m, g( y) < .
Proof. Fix M such that M N of cardinality < .
1) First note that
()
1
if D is an ultralter on M then q(x) is weakly orthogonal to r
D
= Av(D, N).
[Why? As r
D
does not split over M, by 1.4(1).]
Second, note that
()
2
the following type cannot be nitely satisable in M:
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

r
(y) = (x
1
)((x
1
,

d) (x
1
, y)) : (x,

d) q
(x
2
)((x
2
,

d) (x
2
, y)) : (x,

d) q.
[Why? Otherwise for some ultralter D on M we have (y,

d) r
(y) (M,

d)
D. Let b C realize Av(D, N) so as q( x) is closed under conjunction, q(x)(x, b)
and q(x)(x, b) are nitely satisable in C, and we get a contradiction to ()
1
.]
()
3
there is (x,

d) q such that (x
1
)((x
1
,

d) (x
1
, y)), (x
2
)((x
2
,

d)
(x
2
, y)) is satised by no b M.
[Why? By the monotonicity in (x,

d) and q being closed under conjunctions this
follows from ()
2
.]
()
4
let
(x,y),M
(x,

d
(x,y),M
) = (x,

d), from ()
3
()
5
for every b M we have N [= (x)((x,

d) (x, b)) or N [= (x)((x,

d)
(x,
b)).
[Why? By logic this follows by ()
3
.]
()
6
there is
M
2 such that for every b M we have [= (x)((x,

d)(x, b))
i [= (x)((x, y)) (x, b) i (b) = 1.
[Why? By ()
5
+ ()
3
.]
So we are done.
2) Similarly.
1.15
{3n.5}
Claim 1.16. 1) In the previous claim 1.15, if cf() > [T[ then does not depend
on M though

d in general does, i.e. given ( x, y) we may assume without loss of
generality that
( x,

d) =
( x, y)
(x,

d
( x, y),M
).
2) Assume
2
N,q,
from claim 1.13, i.e. N is -saturated and q S
nsp,m
(N). Then
the following partial order is -directed
(a) elements: q B for B N of cardinality
4
[T[
(b) order: p
1
p
2
if p
2
p
1
.
Proof. 1) As if N
1
N
2
N and |N
2
| < , then
( x, y),N2
(x,

d
( x, y),N2
) can
serve as
( x, y),N1
( x,

d
( x,q),N2
).
2) Easy.
1.16
As a conclusion we can now show that a key fact in [Sh:877] for the theory
T = Th(Q, <) has a parallel for every dependent T.
{3n.6}
Conclusion 1.17. The Saturated Expansion Conclusion Assume
(a) N is -saturated
(b) [A[ <
(c) if a A then q
a
= tp( a, N) S
nsp,<
(N), see Denition 1.11(3)

(d) N
[A]
has elimination of quantiers.
Then N
[A]
is -saturated.
4
from some form of strongly dependent we should be able to get essentially nite
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

16 SAHARON SHELAH
Remark 1.18. 1) Recall N
[A]
is N expanded by R
( x,
b)
= a
g( x)
N : C [= [ a,
b]
for ( x,
b) a formula with parameters from A, see [Sh:783, 1].

2) We can omit assumption (d) in 1.17, but then get -saturated only for quantier
free types.
Proof. Without loss of generality is regular, this as it is enough to prove
+
-
saturation for every [[A[, ]. Let M N be such that |M| < and assume
p = p( y) S
m
([M[, N
[A]
) and we shall prove that some c
m
N realizes p( y).
Actually without loss of generality M
[A]
N
[A]
and by assumption (d), equivalently
p( y) S
m
(MA) = S
m
(MA, C) is nitely satisable in M. Let c = c
: <
)
list A so
< and for u
let c
u
= c
: u), x
u
= x
: u).
Next note that by Claims 1.16(1) and 1.15(2) (here clause (c) of the assumption
is used) applied to tp( c
u
, N) noting y of length m is xed and letting x
u
= x
:
u), we have:
()
1
for every nite u
and formula = ( x
u
, y, z) L(
T
) there are
( xu, y, z)
( x
u
,

d
( xu, y, z),M
) tp(c
: u), N) so

d
( xu, y, z),M

>
N
and a function from
g( y)+g( z)
M to 0, 1 such that
( xu, y, z)
( x
u
,

d
( xu, y, z),M
) ( x
u
,
b, c)
(
b c)
:

b
g( y)
M and c
g( z)
M.
Clearly [p( y)[ < so there are
< and a sequence (
( x
u
, y, z
), u
) : <
)
listing the pairs (( x
u
, y, z), u) as above so we have
p( y) =
(c u
, y, c) : <
< and c
g( z
)
M
so u
is nite.
For each <
we choose
( x
u
,

d
) as guaranteed by ()
1
above (for
( x
u
, y, z)).
Let
p
( y) := ( x
u
)[
( x
u
,

d
( x
u
, y, c)] : <
and c
g( z
)
M.
Now
()
2
p
( y) is nitely satisable in M.
[Why? As p( y) nitely satisable in M, using the translation and the choice of
( x
u
,

d
). That is, let p
( y) be a nite subset of p
( y) so for some k < ,
=
() <
, c

g( z
)
M for < k we have p
( y) = ( x
u
()
[
()
( x
u
()
,

d
()
)
( x
u
()
, y, c
)] : < k. Now
(c u
, y, c
) : < k is a nite subset of p( y)

hence is realized by some

b
m
M, hence by the ()
1
the sequence

b realizes p
( y).]
()
3
the type p
( y) is over
: <
M N.
[Why? Check.]
()
4
p
( y) has cardinality [A[ +[T[ +|M| < .

[Why? Obvious.]
()
5
there are M
+
and

b such that:
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

(a) M M
+
N
(b) tp(M
+
, M
: <
) is nitely satisable in M
(c)

b M
+
realizing p
( y).
[Why? Easy, e.g. using ultrapower, N is -saturated and ()
3
.]
()
6

b realizes p( y) S
m
(N A) and

b
g( y)
(M
+
)
g( y)
N.
[Why? Follow the translations.]
So we are done.
1.17
{3n.7}
Question 1.19. 1) Can we waive assumption (d) in 1.17?
2) Is the family of (N, A) as in 1.17 dense under
?
{3n.8.17}
Discussion 1.20. 1) Assume =
<
> = cf() > [T[ and we try to prove that
there is a (, )-limit model.
So let M EC
(T) be saturated and we try to analyze the class of N, M

N EC
, which are close enough, in the sense of (, )-limit model.

So if p S
nsp
<
(M), say p does not split over B, B M of cardinality < , then
we can assume that in N there are enough elements realizing types not-splitting
over B extensions of p. So hopefully we can analyze such N by P S
nsp
(M)
pairwise perpendicular or P S
nsp,|T|
(M) such that for each p P the model

M
[p]
has elimination of quantiers and is saturated, it is reasonable that this holds
if we can expand M by denition of < types p P.
What we need, i.e., what is necessary for this line of attack (but not yet clear if
sucient to carry it)
() if P
above has cardinality and is quite dense (e.g. using Fs for = 1, 2)

then there is an automorphism of M which maps P
1
onto P
2
.
This leads to the generic pair conjecture.
About S
nsp
<
(M) recall Denition 1.11(1).
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

18 SAHARON SHELAH
2. The type decomposition theorem
{tp.45.8}
Context 2.1. 1) T is a complete rst order theory; dependent if not said otherwise.
2) C = C
T
is a monster for T, etc. as in 0.1.
Here we try to analyze a type p S
(M) for -saturated M C where >

[T[, the characteristic case being >> ( may be
0
, if T is countable). In
the case of < [T[, or even better <
0
we know less but mention it. We look
at T being stable as our dream, our paradise. The hard reality is T being just
dependent. In some sense T dependent should be like stable but we allow order,
e.g. Th(Q, <) or trees. What we actually do is investigate the K
(see Denition
2.6).
How helpful is this analysis? We present two consequences. The rst to some
extent accomplished the professed aim: the Type Decomposition Theorem 2.4.
What is its meaning? If M is -saturated,

d

+
>
C and > [T[ then try
to analyze the type tp(
d, M) in two steps: for some c, B

(a) B M has cardinality < , say B = [N[
(b) the type in the rst step is similar to the types of stable theories, i.e.
tp( c, M) does not split over B M; (we can even demand tp( c, M) is
nitely satisable in B)
(c) the type in the second step, tp(
d, M + c) behaves as in trees; e.g. letting

x = (M, B, c, d) we have: on

M the partial orders
x
is -directed where
we let a
1

x
a
2
i tp(
d, c a
2
) tp(
d, c a
1
).
The reader may say that clause (b) is not a true parallel to a stable case, as [B[
is not bounded by + [T[. Still a type not splitting over a set is a weak form of
denability. Also we may wonder, what is the meaning, when T is Th(Q, <)? If
M is -saturated each p S(M) actually stands for a cut of M. Now the cuts are
divided to those which have conality from both sides (falling under (c)), and
those which do not (hence fall under (b)).
The second consequence deals with singular of conality > [T[. Assume M C
is exactly -saturated, i.e. is -saturated but not
+
-saturated. For e.g. T =
Th(Q, <) this is impossible, for T stable there is no problem to nd such M, e.g.
cf() > [T[ and let M be -prime over an indiscernible set of cardinality . The
result says that for dependent T there is something like that, this is 2.2.
{tp16.14}
Lemma 2.2. Singular Exact Saturation Lemma Assume is singular of conality
> [T[ and M C is an exactly -saturated model, i.e. is -saturated but not
+
-saturated. Then there are N and A such that
(a) N M of cardinality < and A M of cardinality and M omits some
p S(A) which does not split over N; in fact
(b) there is q S(M) which does not split over N such that p = q A
(c) there is an indiscernible sequence a
: < ) over N of -tuples such

that Av( a
: < ), N a
: < ) is omitted by M (equivalently, we

cannot choose a
M) and does not split over N.

Remark 2.3. We can add in (c):
(c)
+
moreover we can demand that there is an ultralter D on N such that M
omits p = Av(D, A) where A := a
: < N.
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

{tp25.43}
Theorem 2.4. The Type Decomposition Theorem Assume cf() > [T[, M is
-saturated and

d

C. Then for some c

C, recalling Denition 1.11 we have
tp(c, M) S
nsp,
<
(M) and (P,
P
) is a -directed partial order where:
(a) P = tp(
d, A c) : A M has cardinality
(b) p
1

P
p
2
i p
2
p
1
.
Remark 2.5. 1) In fact (P,
P
) is (P
x,
,
x,
) from the Denition 2.6(1B) below.
2) Note that being
+
-directed is obvious.
3) Would it be more transparent to use the notation p
2
p
1
instead of p
1

P
p
2
?
A matter of taste, the author feels that not.
{tp14.21}
Denition 2.6. 1) Let K = K
1
be the family of x satisfying
(a) x = (A, B, c,

d) but if A = [M[, as usual, we may write M instead of A
and if B = we may omit it
(b) B A
(c) I a linear order
(d) c = c
t,n
: n < n
t
, t I) where n
t
each c
t,n
a nite
5
sequence and let
c
t
= c
t,0
c
t,1
. . . c
t,nt1
(e) c
t,n
: n < n
t
) is a indiscernible sequence over A c
s
: s It, so if
n
t
= 1 this is an empty statement
(f)
1
if t I then
6
tp( c
t
, c
s
: s <
I
t A) does not split over B
2) Let K
0
be dened similarly omitting clause (f)
1
.
3) For , cf() (or just ), cf() and 0, 1 let K
,,<
=
(M, B, c,

d) K
: M is -saturated, [B[ < and [g(
d)[ +[g(c)[[ < ; omitting

means 1. If =
+
instead of < we may write .
4)
(a) let x = (A
x
, B
x
, c
x
, d
x
) for x K
0
(or M
x
instead of A
x
), I = I
x
, c
x,t
=
c
t
, n
t
= n
x,t
and C
x
= Rang( c
t,n
) : t I and n < n
x,t
(b) we may
7
replace

d by D
x
= Rang(
d)
(c) we may omit if = .
For =
0
, let < mean A
x
is the universe of M
x
C (no saturation demand).
5) We dene a two-place relation
0
on K
0
: x
0
y i A
x
A
y
, B
x
B
y
, I
x

I
y
, c
x
= c
y
I
x
, i.e. t I
x
c
y,t
= c
x,t
moreover t I
x
n
y,t
= n
x,t
and
t I
x
n < n
x,t
c
y,t,n
= c
x,t,n
,

d
x

d
y
and tp(c
x
, A
y
x
hence tp(c
x
, A
x
x
follows.
5A) x
1
y mean x
0
y and A
y
= A
x
.
6) We dene a two-place relation
2
on K
0
: x
2
y i x
1
y

d
x
=

d
y
.
7) x K
0
,,
is called normal when Rang(c
x
) Rang(
d
x
).
8) For x K, let P
x,
= (P
x,

x,
) be dened by
5
we remark when it matters
6
of course, this implies clause (e) but see part (2)
7
we may write Cx = Rang(c), cx,t,n = c
x
t,n,m
: m < g(cx,t,n) so in fact cx = c
x
t,n,m
:
(t, m, m) J for the natural J = Jx.
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

20 SAHARON SHELAH
(a) P
x
= tp(
d
x
, A + c
x
) : A A
x
has cardinality
(b)
x,
is the following two-place relation on P
x,
: p
1
( x
dx
)
x,
p
2
( x
dx
) i
p
2
p
1
.
9) If = [T[ +[g(
d
x
)[ +[g(c
x
)[, i.e. we use <
+
, we may omit it.
{tp14.28}
Claim 2.7. 1) K
1
K
0
.
2)
i
is a partial order on K
0
for i = 0, 1, 2.
3) If i 0, 1, 2, x
: < ) is
i
-increasing in K
0
,,
where is a limit ordinal,
[cf()
+
j<
c
xj
= c
x0
] and
8
[i 1 cf()
+
(
<
d
x
=

d
x0
)], [i =
0 cf()] and < cf() (
<
B
x
= B
xi
), then it has a
i
-lub x
:=
x
: < K
0
,,
dened by A
x
= A
x
: < , B
x
= B
x
: <
, I
x
= I
x
: < , c
x
= c
x
: < , i.e. c
x,t
= c
x,t
when t I
x
and
d
x
=
d
x
: < .
3A) In part (3), if < x
K
1
,,
then x
K
1
,,
.
4) If

d

+
>
C and M is -saturated and > then x = (M, , <>,

d)) K
i
,
for i = 0, 1.
5) In the denition of x K
1
: in clause (e) it suces to demand that: if n
t
> 1
then c
t,n
: n < n
t
) is indiscernible over A c
s,m
: s <
I
t, m < n
s
.
6) For every x K
,,
there is a normal y K
,,
satisfying x
1
y, c
x
= c
y
and Rang(
d
y
) = Rang(
d
x
) Rang(c
x
). Hence y mxK
,,
x mxK
,,
, see
Denition 2.8 below.
7) If i = 0 and x
: < ) is
i
-increasing in K
0
,,
and cf() <
+
, < cf(),
then the sequence has a
i
-upper bound x
K
0
,,
, note that we have not said
lub.
7A) In part (7), if < x
K
1
,,
then we can add x
K
1
,,
.
Proof. Easy e.g.
7), 7A) The problem is when part (3) does not cover it, so >
0
. It is clear how
to choose c
x
,

d
x
and B
x
, but we should choose a -saturated M

x
.
Let B = B
x
: < , I = I
x
: < 0 and c = c
t
: t I) with c
t
= c
x,t
when I
x
; similarly d.
First, choose a -saturated M extending M
x
: < but what about
tp( c
x
, M) does not split over B
x
for each < ?
Now for each < < , tp(c
x
, M

x
hence p
,
( x) =
tp(c
x
, M

x
hence p
( x) := p
,
( x) : (, ) =
tp(c
x
,
<
M

x
for the appropriate x. Also for < by
1.2(1) there is c
such that tp(c
, M) does not split over B

x
and extends p
. As
M
x
B
x
is -saturated hence -saturated. Recalling 1.2(1), by the last two
sentences c
I
x
realizes tp(c
, M) for < < hence without loss of generality

< < c
I
x
= c
.
Hence there is an elementary mapping f mapping with domain c
: <
M, mapping c
to c
x
for < , and extending id
{M:<}
. Now M
x
:= f(M)
will do, i.e. let M
y
= M, B
y
= B, c
y
= c,

d
y
=

d.
2.7
8
this follows from x : < is
i
-increasing when i = 2
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

{tp25.32}
Denition 2.8. 1) For = 0, 1 let mxK
,,<
be the family of x K
,,<
which
are
+
2
-maximal in K
,,<
, i.e. for no y do we have x <
+
2
y K
,,<
, see below;
if = 1 we may omit it.
2) For = 0, 1, 2 let
+
be the following two-place relation on K

0
: x
+
i
y i
x
i
y, see Denition 2.6 and if x ,= y then for some t I
y
I
x
satisfying n
x,t
2
we have: C [= [
d
x
, c
t,1
,
b] [
d
x
, c
t,0
,
b] for some =
t
( x, y, z) L(
T
) and
b A
x

c
y,s
: s I
y
t.
3) Again, if =
+
instead of < : we may write , and if = we may omit .
4) Of course, x <
+
i
y means x
+
i
y x ,= y.
{tp25.31}
Observation 2.9. Let i = 0, 1, 2.
1) For = 0, 1 the two-place relation
+
i
is a partial order on K
.
2) If x
1

i
x
2
<
+
i
x
3

i
x
4
then x
1
<
+
i
x
4
.
3) If x <
+
i
z are from K
,,<
then there is y K
,,<
such that x <
+
2
y
i
z
and

d
y
=

d
x
, I
y
I
x
is nite.
Proof. E.g.
3) Let t I
z
I
x
and = ( x, y, z) L(
T
) and

b A
x

c
y,s
: s I
y
t
be such that C [= [
d
x
, c
y,t,1
,
b] [
d
x
, c
y,t,0
,
b]. We choose a nite I I

y
t
such that

b c
s
: s I A
x
. Now dene y by: M
y
= M
x
, I
y
= I
x
I, c
y
=
c
z
I
y
,

d
y
=

d
x
and B
y
= B
x
.
Now check.
2.9
The following claim may be good for digesting the meaning mxK
,,
.
{tp25.30}
Claim 2.10. The L.S.T. Claim for mxK
If x mxK
,
and M = M
x
then for some function F with domain [M]
<
satisfying F(A) [M]
2
|A|+|T|
for A Dom(F), we have: if M
1
M is closed
under F and contains B
x
then (M
1
, B
x
, c
x
,

d
x
) belongs to mxK
,
.
Proof. We can choose F() [M]
1
and for notational transparency we x a set J
of cardinality
0
disjoint to I
x
.
Note that for every N M satisfying B
x
N we have x
N
:= (N, B
x
, c
x
,

d
x
)
K
,
; call such N a candidate. So to choose F let us analyze the cases B
x
N M
but x
N
/ mxK
,
. Considering Denition 2.8, it suces by 2.9(3) to consider the
case x
N
<
+
2
y, I
y
I
x
is nite and is J, t
I
y
I
x
,
( x, y, z),
as there, we can
ignore the possibility that also some other t I
y
I
x
t
works.
We let
b
0
list B
y
, I
y
I
x
= t
0
, . . . , t
n1
, t
= t
()
, () < k, m
t
> g( c
y,t,0
), n
t
=
n
y,t
for t I
y
. Also without loss of generality = ( x, y; z
1
, z
2
), g( y) = g( c
y,t,0
), g( x) =
g(
d
x
),
b =

b
1
b
2
,
b
1

>
(M
y
) =
>
N, g( z
1
) = g(
b
1
) and abusing our notation
b
2
= c
y,t,k
: k < n
y,t
, t I
y
t
), so C [= [
d
x
, c
y,t,1
,
b
1
,
b
2
][
d
x
, c
y,t,0
,
b
1
,
b
2
]
and let g( z
t
) = g( c
y,t
), g( z
t,
) = g( c
y,t,
) so z
t
= z
t,0
. . . z
t,nt1
and z
2
=
(. . . , z
t
, . . .)
tIy
.
All this information will be called a witness against the candidate N and we
denote it by w.
Let s consist of the following pieces of information on the witness w and we shall
say that w materializes s and s is a case for N.
(a) I = I
y
and t
: < k) (so we will write I

s
, t
s,
)
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

22 SAHARON SHELAH
(b) n
t
, m
t
for t I
(c) = ( x, y, z
1
, z
2
) hence g(
b
1
)
(d) (0) = g(
b
0
)
(e) q
0
= tp(
b
0
, ) and q
1
= tp(
b
0
b, )
(f) the scheme of non-splitting of tp( c
t
, M) for t I
y
I
x
from
clause (f)
1
of 2.6(1), that is
t
= (( z
ct
, y
), q( y
, y
b0
)): for some
b
g( y
)
N
we have C [= [c
t
,
b] and q( y
, y
b0
) = tp(
b
0
, ).
We shall write t
= t
(s), =
s
( x
s
, y, z
s,1
, z
s,2
), q
s,0
= q
0
, etc. and let r
s
=
tp( c
y
, N). We call s a case when it is a case for some candidate N. If s is a case
and
b
0

(0)
(M
x
) realizes q
s,0
, then we can choose c
s,
b0
= c
x,
b0,t
: t I
s
) such that
tp( c
s,
b0,t
, c
s,
b0,s
: s <
Is
t M
x
) is dened by the scheme
t
with the parameter
b
0
, this type is determined by s and x (though not the c
s,
b0,t
s themselves). Without
loss of generality t I
x
c
s,
b0,t
= c
x,t
.
Now clearly
()
1
for a candidate N we have x
N
K
,
i for every case s and

b
0

(0)
N
realizing q
s,0
there is no

b
i

g( y)
N such that s
1
,
b
0
,
b
1
,
b
2
= c
s,
b0
are as
above.
So
()
2
for every case s,
b
0

(0)
(M
x
) realizing q
s,0
and

b
1

g( zs,1)
M, we cannot
choose c
, c
realizing tp(c
s,
b0
, M
x
) such that c
I
x
= c
x
= c
I
x
and C [=
s
[d
x
, c
t(s)
, c
(I
s
t
(s)),
b
1
,
b
2
]
s
[d
x
, c
t(s)
, c
(I
x
t
(s)),
b
1
,
b
2
].
[Why? As then x / mxK
,
.]
Hence
()
3
for every case s and

b
0

(0)
(M
x
) realizing q
s,0
and

b
1

g( zs,1)
(M
x
) there
is a nite set C = C
s
(
b
0
,
b
1
) M
x
such that: if N is a candidate which
includes

b
0
,
b
1
, C then there is no witness w against N with s = s
w
,
b
0
=
b
w,0
,
b
1
=

b
w,1
.
Also
()
4
for B M of cardinality < let C(B) be a subset of M of cardinality
2
|B|+|T|
in which every p S
<
(B) is realized.
Lastly, let F be dened for B [M]
<
, F(B) = C
s
(
b
0
,
b
1
) : s a case and

b
0
,
b
1
suitable sequences from B C(B).
Now the number of cases is
0
+
0
+
0
+ [T[ + + 2
|T|+|A|
= 2
|T|+|A|
, so
F(B) [M]
. So we are done.
2.10
{tp25.33}
Theorem 2.11. The Existence Theorem If = 0, 1 and cf() > [T[ and
x K
,,
then there is y such that x
2
y mxK
,,
.
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

{tp25.34}
Remark 2.12. 1) If we use K
,,<
instead of [T[ we should demand cf() >
[T[.
2) We may get more. E.g. demand I
y
= I
1
I
2
, I
1
< I
2
, i.e. s
1
I
1
s
2
I
2

s
1
<
I
s
2
and I
1
= I
x
, I
2
is well ordered.
2A) Also this claim holds (by the same proof) when we replace clause (f)
1
in
Denition 2.6(1) by
(f)
2
tp(c
x
, A
x
) is nitely satisable in B
x
.
Then in part (2) of the remark we may add
() for t I
2
, tp( c
t
, M
x
c
y,s
: s <
Iy
t) is nitely satisable in B
y
.
See more in 2.32, 2.33 and 2.34.
3) We can be more relaxed in the demands on x
: <
+
) in the proof e.g. it
suces to demand
(a) x
K
0
,,
(b) x
is
1
-increasing continuous, natural to demand

d
x
=

d
x0
but not
necessary, and
(c) for each <
+
(or just for stationarily many <
+
) we have
x
<
+
1
x
+1
.
[Why? Let

d
x1
u
(u
nite) t
( x
, y
) witness x
<
+
1
x
+1
when S
0
where S
0
:= <
+
: x
<
+
1
x
+1
. For S let h() = Min :

d
x
u
d
x
u
, equivalently u
Dom(
d
x
) and Rang(
) (A
x
+ c
x
) A
x
+ c
x
,
clearly h() < for limit S
0
.
So by Fodor Lemma for some < and u the set S = S
0
: is a limit
ordinal as in clause (c) above u
= u and h() = is stationary. As [T[, for

some
9
formula the set S
2
:= S
1
:
= is a stationary subset of
+
and
we continue as in the proof.
4) How does part (3) of the remark help? E.g. if we like to get y mxK
,,
which is normal and Rang(
d
y
) is universe of some N C.
Proof. Assume this fails. We try to choose x
by induction on <
+
such that
(a) x
,,
(b) x

1
x
for <
(c) if = + 1 then x
<
+
2
x
, i.e. x
witness x
/ mxK
,,
(i.e. is like y in Denition 2.8);
(d) x
0
= x.
For = 0 use clause (d), for = +1 we use our assumption toward contradiction.
For limit use 2.7(3). Note that

d
x
=

d
x
for <
+
by clauses (d) + (b).
Having carried the induction, for each <
+
there are t
( x, y
, z
),
satisfying:
()
1
(a) t
I
x+1
I
x
9
if ct,n is innite we let u g( c
x
t
,0
) be nite such that C |= [
dx, c
x
+1
t
,0
u,
b]
[
dx
, c
x
+1
t,1
,
b] and the rest is the same.

(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

24 SAHARON SHELAH
(b)
( x, y
, z
) L(
T
)
(c)

b
M
x

c
x+1,s
: s I
x+1
t
(d) g(
) = g( z
), g( y
) = g( c
x+1,t,0
)
(e) C [=
d
x
, c
x+1,t,0
,
d
x
, c
x+1,t,1
,
] so g( x) = g(
d
x
)
and n
t
2.
Clearly the sequence t
: <
+
) is without repetitions. Now let J
I
x+1
t
be nite such that
M
x
c
x+1,s
: s J
. We can nd a pair (( x, y, z), n
)
such that the set S
0
= <
+
: [J
[ n
and
( x, y
0
, z
0
) = ( x, y, z) is innite
(even stationary). By Ramsey theorem (or Fodor Lemma) we can nd an innite
(and even stationary) set S
1
S
0

+
such that S
1
J
: S
1
= .
Note that there are possibilities for , not necessarily [T[ because though
( x, y
, z
) depend only on x u for some nite u g( x) there are g(
d
x
)+
0
possibilities for u.
Next we shall prove that in this case ( x; y, z) has the independence property
(for T), a contradiction.
For every w v S
1
and
w
2 let, noting that u = Dom():
(a) A
,v
= A
x
c
x
+1
,t,n
: t J
and n < n
x
+1
,t
for some v
c
x+1,t,()
: u
(b) f
,v
is the function with
() domain A
,v
() is the identity on A
x
c
x
+1
,t,n
: t J
and n < n
x
+1
,t
for some
v
() f
,v
( c
x+1,t,()
) = c
x+1,t,0
for w = Dom().
Now
() f
,v
is an elementary mapping.
[Why? Without loss of generality w v are nite so () for some () <
+
and prove this by induction on [u[. We just use: for u the sequence c
x
d()
,t,n
:
n < n
x
()
,t
) is indiscernible over A
x
()
c
x
()
,s
: s I
x
()
t, by Denition
2.6(1), clause (e).]
Now let g
,v
aut(C) extends f
,v
. So u g
,v
(
) = f
) =

b
and
g
,
( c
x+1,t,()
) = c
x+1,t,0
for w; hence by the choice of J
so for
w
2
we have
() C [= [g
,v
(
d
x
), c
x+1,t,0
,
] i () = 1.
So ( x, c
x+1,t,0
,
) : t u) is an independent sequence of formulas; as w is any

subset of S
1
we get a contradiction as promised.
2.11
{tp25.36}
Claim 2.13. The Weak Orthogonality Claim
Assume = 0, 1 and x mxK
,,
.
1) If m < , B
M
x
and
10
[B
[ < and q S
m
(M
x
C
x
then tp(
d
x
, M
x
C
x
) is weakly orthogonal to q.
10
this is just to ensure that M realizes every q S
<
(B
)
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

1A) It suces
11
that q = tp( c
1
c
2
, M
x
C
x
), tp( c
1
, M
x
) does not split over some
B
[M]
<
, tp(C
x
, M
x
+ c
1
x
and tp( c
2
, M
x
+ c
1
+C
x
) does
not split over some B
[M
x
]
<
.
2) If x
mxK
,,
for < is
1
-increasing, <
+
and x
:= x
: <
belongs to K
,,
(recall 2.7(3)) then it also belongs to mxK
,,
.
Proof. 1) Assume toward contradiction that those types are not weakly orthogonal.
Let q = q( y), y = y
k
: k < m) and let x = x
: <
x
) recalling
x
= g(
d
x
) and
p( x) = tp(
d
x
, M
x
C
x
). So for some formula ( x, y, z) and e
g( z)
(M
x
C
x
) the
type p( x) q( y) does not decide ( x, y, e), i.e. r
t
( x, y) = p( x) q( y) ( x, y, e)
t
is consistent (= nitely satisable in C) for t = 0, 1 and let c
0
, c
1
be such that

d c
t
realizes r
t
( x, y) for t = 0, 1. Now it cannot be that tp( c
t
, M
x
C
x

d
x
) does not
split over B
for both t = 0, 1 (by 1.2(1), as every p S

<
(B
) is realized in M
recalling M is -saturated and > [B
[). So choose c
0
c
0
, c
1
such that the
type tp( c
0
, M
x
C
x

d) splits over B
.
Now by 1.2(1) there is c
1

m
C such that tp( c
1
, M
x
C
x
d
x
c
0
) extends q( y)
and does not split over B
. Hence also tp( c

1
, M
x
C
x

d
x
) does not split over
B
, hence it is dierent from tp( c

0
, M
x
C
x

d
x
). We can continue and choose
c
n
(n = 2, 3 . . .) realizing the complete type over M
x
C
x

d
x

c
k
: k < n
which extends q and does not split over B
. Hence
()
0
for every n < , tp( c
n
, M
x
C
x
c
k
: k < n) extend q( y) and does not
split over B
.
So by 1.2(2)
()
1
c
n
: n < ) is an indiscernible sequence over M
x
C
x
.
Also (by induction on ) by 1.2(3) we have:
()
2
if t I
x
then c
x,t,n
: n < n
t
) is an indiscernible sequence over M
x
c
x,s,m
:
s It c
n
: n < .
Now we dene y = (M
y
, B
y
, c
y
,

d
y
) by
(a) M
y
= M
x
(b) B
y
= B
x
B
(c)

d
y
=

d
x
(d) I
y
= I
x
s() were [t I
x
t <
Iy
s()]
(e) c
y
I
x
= c
x
(f) n
y,s()
= (or any number [2, ]) and c
y,s(),n
= c
n
.
Now y contradicts the assumption x mxK
,,
.
1A) Similarly.
2) Easy.
2.13
The following claim is a crucial step toward proving the Type Decomposition
Theorem 2.4.
11
even more as we can increase the linear order Ix
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

26 SAHARON SHELAH
{tp25.38}
Claim 2.14. If x mxK
,,
(or just x K and below) then for every A M
x
of cardinality < and = ( x, y, z) L(
T
) satisfying g( x) = g(
d
x
), g( z) =
g( c
x
), g( y) = m, there is ( x, e, c
x
) tp(
d
x
, M
x
C
x
) satisfying e
>
M
x
such that ( x, e, c
x
) ( x,
b, c
x
)
t
:

b
g( y)
A and t 0, 1 are such that C [=
[
d
x
,
b, c
x
]
t
where
if q( y) S
g( y)
(M
x
c
x
) is nitely satisable in some A M
x
of cardinality
< then q( y) is weakly orthogonal to tp(
d
x
, M
x
C
x
).
Proof. By 2.13, if x mxK
,,
then holds; hence we can in any case assume .
Let p( x) = tp(
d
x
, M C
x
), so x = x
i
: i < g(
d
x
)) and recalling y = y
: <
m) we dene a set r = r( y) as follows
r( y) := ( x)(( x, y, c
x
)
t
( x, a, c
x
)) : t 0, 1 and
( x, a, c
x
) p( x) and a
>
(M
x
).
Now
1
r( y) is not nitely satisable in
m
A.
[Why? If r( y) is nitely satisable in
m
A, then there is an ultralter D on
m
A
such that for every ( y, a, c
x
) r( y), the set
b :

b
m
A and [= [
b, a, c
x
]
belongs to D. Let q( y) = Av(D, M
x
c
x
), clearly q( y) S
m
(M
x
c
x
) is nitely
satisable in
m
A, A M and [A[ < . Let

b
m
C realize q( y), so (( x, a, c
x
)
p( x)) t 0, 1 C [= ( x)(( x,
, c
x
)
t
( x, a, c
x
)). Why? This holds by
the choices of r( y), D and

b
. As p( x) is closed under conjunctions it follows that

p( x) ( x,
, c
x
)
t
is nitely satisable in C for t = 0, 1. But this contradicts the
assumption .]
Hence for some n < and
( x, a
, c
x
) p( x) for < n we have
2
for no

b
m
A do we have ( x)(( x,
b, c
x
)
t

( x, a
, c
x
)) for < n, t
0, 1.
Let ( x, a, c
x
) =
<n
( x, a
, c
x
), so clearly
3
(a) a
>
(M
x
)
(b) ( x, a, c
x
) p( x)
(c) for no

b
m
A do we have
1
t=0
( x)(( x,
b, c
x
)
t
( x, a, c
x
)).
So for every

b
m
A for some t = t(
b) 0, 1 we have C [= ( x)(( x,
b, c
x
)
t
( x, a, c
x
)) hence ( x, a, c
x
) ( x,
b, c
x
)
t(
b)
. As ( x, a, c
x
) p( x) = tp(
d
x
, M
x
c
x
) it follows that ( x,
b, c
x
)
t(
b)
p( x).
So ( x, a, c
x
) is as required.
2.14
{tp14.42}
Claim 2.15. Assume x mxK
,,
. If B
M
x
, [B
[ < ,

d
m
(D
x
), c
>
(C
x
), = ( x, y, z), g( x) = g(
d), g( z) = g( c) then for some ( x, y
, z
) and
e
g( y
)
(M
x
) we have [= [
d, e, c] and
(x, e, c) ( x,
b, c)
t
:

b
g( y)
(B
) and C [= [
d,
b, c]
t
.
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

Proof. This just reformulates 2.14; see more in 2.19(0).
2.15
Now at last
Proof. Proof of 2.4, The Type Decomposition Theorem
By 2.7(4) there is x K
1
,
such that d
x
=

d and M
x
= M. By Claim 2.11
without loss of generality x mxK
1
,
. Clearly (P
x,
,
x,
) is a partial order.
Assume that () < and p
P
x,
for < (). Let B =
Dom(p
) : <
() B
x
, so B M
x
has cardinality < . Hence by 2.15 for every v g(c
x
)
and = ( x, y, z) satisfying g( z) = g(c
x
), g( x) = g(
d
x
), g( y) < there is
=
( x, e
, c
x
) tp(
d
x
, M c
x
) where e

g( y
)
(M
x
) such that
( x, e
, c
x
)
( x,
b, c
x
)
t
:

b
g( y)
B and C [= [
d
x
,
b, c
x
]
t
.
Let A = Rang( e
) : as above, and let p = tp(
d
x
, A c
x
); it is an upper
bound, as required.
2.4
{tp14.44}
Discussion 2.16. The type decomposition theorems say that we can analyze a type
p S(M) in two steps; rst tp( c, M) does not split over some small B M.
Second, tp(
d, c + M) is like a type in the theory of trees and lastly without loss

of generality by 2.7(6) some initial segment of

d realizes p. As an example, see
[Sh:877]:
{tp14.45}
Exercise 2.17. Let T = T
ord
be Th(Q, <) and if M C
T
, M is -saturated, in
the main case and p S(M) then
(a) p induces a cut

C
p
of M where

C
p
= C
p,1
, C
p,2
), C
p,1
= a M : (a <
x) p and C
p,2
= MC
p,1
(b) now

C
p
has a pair (
1
,
2
) of conalities, that is
1
= cf(C
p,1
, <
M
),
2
=
cf(C
p,2
, >
M
) and max
1
,
2

(c) now p does not split over some subset B of M of cardinality < i
min
1
,
2

(d) for every B M of cardinality < min
1
,
2
for some (x, a) p we
have (x, a) p B (i.e. p under is min
1
,
2
-directed)
(e) so for = min
1
,
2
we have a decomposition which is trivial in some
sense: either we have c =<> or we have

d = c
(f) if e.g.
1
<
2
and B
1
is an unbounded subset of C
p,1
of cardinality
1
and
c, d realize in C (where M C) the type p and C [= c < d, then tp(c, M) is
nitely satisable in B
1
and for every A M of cardinality <
2
for some
formula (x, a) p we have (x, a) (c < x) tp(d, A c).
{tp14.46}
Discussion 2.18. Note: if T is stable and x mxK
,
is normal then

d
x

dcl(Rang(c
x
)).
{tp26.7}
Claim 2.19. 0) In 2.15 if cf() > +[T[ then we can choose = ( x, y
, z
) such
that it depends on x,

d, ( x, y, z) but not on B
.
0A) If cf() > 2
+|T|
then we can x also q = tp( e, C
x
d
x
). If cf() > 2
+|T|+|Bx|
then we can x q = tp( e, B
x
C
x

d
x
).
1) Assume that x K
0
and = ( x, y, z) L(
T
) are such that y, z are nite and
g( x) = g(
d
x
).
Then the following set is nite: J = J
x,
= t I
x
: there are =
: <
g( y)) and a sequence

b
g( y)
(A
x
c
x,s
: s I
x
t) such that
g( c
x,t,0
)
and C [= [
d
x
, ( c
x,t,0
)
: < g( y)),
b] [
d
x
, ( c
x,t,1
)
: < g( y)),
b)].
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

28 SAHARON SHELAH
2) Moreover, the bound depends just on and T.
3) For any x K
<0,<0
and there is y satisfying x
1
y K
<0,<0
such that
I
y
I
x
is nite and the local version of maximality holds, i.e.
y
1
z K
<0,<0
J
z,
= J
y,
.
4) For any x K
<0,<0
and sequence
n
: n < ) we can nd x
n
: n < )
such that x
0
= x, x
n

1
x
n+1
K
<0,<0
, I
xn
I
x
nite and x
n+1
,
n
satises the
demands on y, above for every n.
{tp26.9}
Conclusion 2.20. If x mxK
,,
and cf() > + [T[ then we can nd e
C and
( x, y, z)
( x, e, c
x
) : ( x, y, z) L(
T
)) satisfying g( x) = g(
d
x
), g( z) =
c
x
, g( y) = such that: for each ( x, y, z) as above we have:
( x, e, c
x
) tp(
d
x
, e c
x
)
( x, e, c
x
) ( x,
b, c
x
)
t
:

b
g( y)
M and t 0, 1 such that C [=
[
d
x
,
b, c
x
]
t
.
Proof. Proof of 2.20
By 2.14 and 2.19 and compactness.
0) As there are + [T[ < cf() possible choices of and the set of possible Bs
is (cf())-directed.
0A) Similarly.
1),2) Let n
1
be minimal such that: for no

b

g( y)
C, c

g( z)
C for < n
1
is
( x,
, c
) : < n
1
) an independent sequence of formulas, exist as T is a dependent
theory. Let n
2
be minimal such that if u
i
[n
2
i]
g( z)
for i < n
2
then for some
v [n
2
]
n1
we have i, j v i / u
j
(the -system lemma, for nite sets, see 0.5).
Now n
2
is a bound as required by the proof of 2.11.
3),4) Follows.
2.19
On strongly/strongly
2
dependent theories see [Sh:863].
{tp26.14}
Claim 2.21. 1) If T is strongly dependent then in the previous claim 2.19(1), if
x satises t I
x
n
t
= then for each n < even the set J
x
= J
x,
:
( x, y, z) L(
T
) as there and g( z) < n is nite.
2) If Y is strongly dependent
2
then above we can allow n = .
Proof. Similar.
2.21

We now turn to exact saturation.
Proof. Proof of 2.2, The Singular Exact Saturation Lemma
Let = [T[. As M is not
+
-saturated, there are A M of cardinality
and p S
1
(A) omitted by M. Let d C realize p. By Theorem 2.4, there is
c

C as there for (d), so in particular such that tp(c, M) does not split over
some N M of cardinality < . Let B
i
: i < cf()) be a -increasing sequence
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

of sets with union N A such that i < cf() [B
i
[ < and N B
0
. Now we
choose A
i
M by induction on i < cf() such that A
i
is of cardinality and
tp(d, A
i
c) tp(d, B
+
i
) where B
+
i
:= B
i

A
j
: j < i.
The choice is possible as [B
+
i
[ < by 2.4, i.e. by the choice of c. Next we can
nd A
M of cardinality such that tp(d, A
c) tp(d,
i<
A
i
), possible as
[

i< cf()
A
i
[ + cf() < . Let B
+
= B
+
i
: i < cf() A
so [B
+
[ = and
we ask the question:
is there an elementary mapping f (or automorphism of C) such that f B
+
is the identity and f(c)

M?
If yes, then let d
M realize f(tp(d, A
f(c)) S(A
f(c)) hence there is an

elementary mapping g, gB
+
= id
B
+ = fB
+
, g(c) = f(c) and g(d) = d
. Hence
easily for each i < cf() the sequence d
)f(c) realizes f(tp(d)c, A

i
)) hence it
realizes also f(tp(d)c, B
+
i
))); so d
realizes f(tp(d, B
+
i
)). But as B
+
i
increases
with i it realizes f(tp(d, B
+
i
: i < cf()), hence it realizes tp(d, A), but d
M
contradicting the choice of p, A, d.
So the answer to the question is no, which gives clauses (a),(b) of the desired
conclusion. As for clause (c), we choose c
by induction on <
() (a) c

M
(b) c
realizes tp(c, c
: < N)
(c) if possible c
does not realize tp(c, N B) hence for some

=
< cf(), c
does not realize tp(c, B
)
(d) if
is well dened, it is minimal.

There is no problem to carrying the induction, by 1.2(2) the sequence c
: )
is indiscernible over N and easily satises the required statement.
Clause (c)
+
, that is moreover, there is an ultralter D on N follows when we
use the version of mxK from 2.12(2A) or 2.32 - 2.34.
2.2
We may have hoped that 2.2 characterize being dependent, but this is not so.
Clarication when this property (characterization of exctly -saturated > cf() >
[T[, as in 2.2) holds is given by
{tp16.15}
Example 2.22. 1) There is an independent T such that: if T has an exactly
-saturated model then is regular. In fact, this is a sucient condition.
2) The same holds for exactly -compact, >
0
.
Proof. We use T which satises The Chang Trick from his proof of his two car-
dinal theorem (
1
,
0
) (
+
, ) when =
<
; the use is not an incident, he uses
such T to overcome a related problem in his proof.
The condition is:
for some predicate R(x, y)
T
written xRy (or just a formula
(x, y)
L(
T
)) we have
12
:
(a) the empty set can be coded, that is yx(xRy)
12
instead (a)+(b) we can have
(a)
(x
0
, . . . , x
n1
)(z)(zRy =
W
<n
z = x
) for every n.
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

30 SAHARON SHELAH
(b) we can add to a coded set one element, that is (x, y)(y
1
)(x
1
)[x
1
Ry
1

(x
1
Ry x
1
= x)].
Note: for any model M, if R /
M
, M an innite model, let u
b
: b M) list the
nite subsets of M, and we expand M to M
+
by choosing R
+
= (a, b) : a u
b
and b M, then Th(M
+
) is as required.
So assume =
i
: i < cf(),
i
<
j
< for i < j < cf(), cf() < and
M is -saturated. Let A M, [A[ = and p S(A, M). Let A = A
i
: i <
cf(), [A
i
[
i
, A
i
increasing with i. Let c
i
realize p A
i
.
By induction on i, we choose b
i
M which realizes the type
p
i
(y) = c
j
Ry : j < cf() and j i (x)(xRy xRb
j
) : j < i
(x)(xRy (x, a)) : ( x, a) p A
i
.
By and the induction hypothesis on i, p
i
(y) is nitely satisable in M.
[Why? Let p
(y) p
i
(y) be nite so it has the form c
j
Ry : j u (x)(xRy
xRb
j
) : j v (x)(xRy
(x, a
)) : < n where u [i, cf()) is nite,

v i is nite and
( x, a
) pA
i
for < n. By we can nd c M such that
M [= (x)(xRc (
jv
x = b
j
), thus c realizes p
(y))].
But [p
i
[ < so we can choose b
i
.
Now xRb
i
: i < cf() is a set of formulas nitely satisable in M of cardinality
< and any element realizing it realizes p.
2.22
A weak complement to 2.4 is 2.25 but rst recall
{tp16.16}
Denition 2.23. [T not necessarily dependent]
1) I

C is a stable indiscernible set when : I is an innite indiscernible set and
Av(I, C) is well dened, i.e. for any ( x, y), g( x) = and

b
g( y)
C either (I,
b)
or (I,
b) is nite.
2) I is a dependent indiscernible sequence when : I is a linear order and I = a
t
:
t I) is an indiscernible sequence and for every formula ( x,
b), g( x) = g( a
t
),
there is a convex equivalent relation E on I with nitely many equivalence classes
such that sEt [ a
t
,
b] [ a
s
,
b].
{tp16.16n}
Fact 2.24. If T is dependent, I

C is a stable indiscernible set i I is an innite
indiscernible set.
Proof. By [Sh:715, 1.28].
{tp16.17}
Claim 2.25. 1) Assume (T is dependent and) there is an innite indiscernible set
I C. If
+
= 2
and > [T[ then T has an exactly -saturated model.

2) Assume (T not necessarily dependent), I C is a stable indiscernible set. Then
the conclusion of part (1) holds.
3) In parts (1),(2) the conclusion holds for T if the assumption holds for T
eq
.
{tp16.18}
Remark 2.26. 1) Of course, trivially if for some non-zero ordinal there is an
innite indiscernible set I

C then for some i < , ( a)
i
: a I is an innite
indiscernible set.
2) But we could use below indiscernible set I

C.
3) On indiscernible sets for T dependent see [Sh:715, 1], we use it freely.
4) Of course: if a
t
: t I

C is an innite indiscernible set/a stable indiscernible
set, u and a
t
u : t I) is not constant then a
t
u : t I
u
C is an
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

innite indiscernible set/a stable indiscernible set. So using singletons in 2.25 is
not a loss.
5) If I is a linear order and I
1
, I
2
I are innite, a
t

C for t I and I = a
:
t I) is an indiscernible sequence then I I
1
is stable i I I
2
is stable.
This is easy.
1) By part (2).
2) Let I = a
: < C be an innite stable indiscernible set. Now easily
1
for any

b
>
C for some J [I]
|T|
, the set IJ is an indiscernible set over
b J.
[Why? By the denition. First prove that for some nite J
we have [= [ a
t0
, . . . , a
tn1
b] =
[ a
s0
, . . . , a
sn
,
b] when t
0
, . . . , t
n1
, s
0
, . . . , s
n
IJ
with no repetitions. Then use

transitivity of equivalence.]
2
the following conditions on

b
>
C are equivalent:
(a) tp(
b, I), Av(I, I) are weakly orthogonal

(b) for some J [I]
|T|
we have tp(
b, J) tp(
b, I).
[Why? Check.]
3
let D = D
I
= p S
<
(I) : p weakly orthogonal to Av(I, I).
We dene
4
we say A is a D-set if a
>
A tp( a, I) D
I
5
if I A we let S
m
D
(A) = tp(
b, A) : A
b is a D-set and g(
b) = m.
We note
6
(a) if A is a D-set and I A then Av(I, I) Av(I, A); hence if A = [M[,
then M is not
+
-saturated
(b) A is a D-set i A I is a D-set
7
if A is a D-set of cardinality < and p S
m
D
(A I) then for some J I
of cardinality [A[ +[T[ we have p (A J) p.
[Why? By
2
and
6
.]
8
if A
: < ) is an -increasing sequence of D-sets, then A
:= A
:
< is a D-set.
[Why? By the denition of a D-set.]
9
if A
: ) is -increasing continuous sequence of D-sets, I A

0
and
p S
m
(A
) then p S
m
D
(A
)

<
p A
S
m
D
(A
).
[Why? By the denition of S
m
D
() and
8
.]
Now comes a major point
10
if A C and [A[ < then we can nd I
1
such that: I
1
is an indiscernible
set, I I
1
, [I
1
I[ [T[ +[A[ and A is a D
I1
-set.
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

32 SAHARON SHELAH
[Why? Let = [A[ +[T[ and we try by induction on <
+
to choose an element
a
of C which realizes Av(I, a
: < I) but does not realize Av(I, a
: <
AI). But if we succeed to carry the induction clearly I
+
:= Ia
: <
+
is an indiscernible set, and a stable one (recalling 2.26(5)) hence for some J I
+
of
cardinality [A[ +[T[, also I
+
J is an indiscernible set over AJ, but necessarily
J I a
: < for some <

+
, easy contradiction to the choice of a
.]
11
if A
1
A
2
are D
I
-sets, [A
2
[ , [A
1
[ < and p S
m
D
(A
1
I) then there
is q S
m
D
(A
2
I) extending p.
[Why
11
? By
9
without loss of generality A
2
= A
1
b, so [A
2
[ < . We can
nd c realizing p( y) and let A = A
1
b c = A
2
c. So by
10
there is I
+
such that:
I
+
is an indiscernible set, I I
+
, [I
+
I[ := [A[ +[T[ = [A
1
[ +[T[ < and A is
a D
I
+-set. As A
2
is a D
I
-set we can nd J
1
I of cardinality [A
2
[ +[T[ <
satisfying tp(A
2
, J
1
) tp(A
2
, I). Also A
1
c is a D
I
-set (as A
1
is a D
I
-set and c
realizes p( y) S
m
D
(A
1
I)) hence there is J
2
I of cardinality [A
1
c[ + [T[ =
< such that tp(A
1
c, J
1
) tp(A
1
c, I). Lastly, as A is a D
I
+-set there is
J
3
I
+
of cardinality [A[ +[T[ = such that tp(A, J
3
) tp(A, I
+
).
As tp(A
2
, J
1
) tp(A
2
, I) necessarily tp(A
2
, J
1
) tp(A
2
, I
+
). Similarly tp(A
1
c, J
2
) tp(A
2
c, I
+
). By cardinality considerations there is a permutation h of
I
+
which is the identity on J
1
, J
2
and J
3
I and maps J
3
I into I. As I
+
is an
indiscernible set, h is an elementary mapping (of C). As h J
1
is the identity
and tp(A
2
, J
1
) tp(A
2
, I
+
), see above also h id
A2
is an elementary mapping
hence there is an automorphism g of C extending h id
A2
. As tp(A
1
c, J
2
)
tp(A
1
c, I
+
) and h J
2
= id
J2
, hA
1
= id
A1
(recalling A
1
A
2
) and h(I
+
) = I
+
necessarily g( c) realizes tp( c, A
1
I
+
) hence it realizes tp( c, A
1
I) which is equal
to p( y). Also tp(A
2
c, J
3
) tp(A
2
c, I
+
) hence tp(A
2
g( c), h(J
3
)) tp(A
2
h( c), I
+
) hence tp(A
2
g( c), h(J
3
)) tp(A
2
g( c), I), but h(J
3
) I. So A
2
g( c)
is a D
I
-set hence q( y) =: tp(g( c), A
2
I) belongs to S
m
DI
(A
2
I) so is as required.]
As 2
=
+
, by
9
+
11
there is M I of cardinality
+
which is -saturated
and is a D-set hence by
6
is not
+
-saturated.
3) Should be clear.
2.25
{tp16.19}
Observation 2.27. (Any complete rst order T)
In C there is no innite indiscernible set i for some n and = (x
0
, . . . , x
n1
)
L(
T
), is connected and anti-symmetric i.e. if a
0
, . . . , a
n1
C with no repetitions
then for some permutations
1
,
2
of 0, . . . , n 1 we have
C [= [a
1(0)
, . . . , a
1(n1)
] [a
2(a)
, . . . , a
2(an1)
].
{tp16.20}
Remark 2.28. 1) The second condition is related to the property (E) of complete
rst order theories of Eherenfeuct [Ehr57] which says that the condition holds for
some innite set.
2) Note that C may have no innite indiscernible set but C
eq
has.
Proof. The implication is obvious.
So assume the rst statement. For and
:= ( x) :
L(
T
), x = x
: < n) let
= y
k
,= y
: k < < (y
k0
, . . . , y
kn1
)
(x
0
, . . . , x
n1
) : n < , (x
0
, . . . , x
n1
) L(
T
) and k
0
, . . . , k
n1
< without
repetitions and
0
, . . . ,
n1
< without repetitions. Easily
is not realized
in C by the present assumption and
k
: is a nite subset of
and k < )
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

is -increasing with k and with union
. Hence for some nite
and
k < , the set
k
is not realized in C.
Let
i
(x
0
, . . . , x
ni1
) : i < i()) list so i() < , so without loss of generality
n
i
< k for i < i(). Lastly, we dene (y
0
, . . . , y
k1
), it says: if y
: < k) is with-
out repetitions and i is minimal such that y
: < k) is not an
i
(x
0
, . . . , x
ni1
)-
indiscernible set then
i
(y
0
, . . . , y
ni1
). Now check.
2.27
{tp16.21}
Question 2.29. 1) Is there a dependent T such that even in C
eq
there is no innite
indiscernible set but some singular of conality > [T[ there is an exactly -
saturated model of T?
2) For a dependent theory T characterize : singular and T has exactly -
saturated models.
3) In both parts we may at least initially restrict ourselves to strong limit of large
enough cardinality such that 2
=
+
.

The following in a sense gives a spectrum for

d/M.
{tp35.46}
Claim 2.30. For [T[, a model M and sequence

d

C, there is a set such
that:
(a)
:= : > and M is -saturated

(b) [[
(c) if and cf() > then there is x mxK
,
such that M
x
= M,

d
x
=
d and [B
x
[ + sup( )
(d) if < and cf() then sup( ) = .
Proof. Straight.
For each
:=
of conality > we can nd x
mxK
,
such that M
x
= M,

d
x
=

d and for
let
be a conal subset of of
cardinality cf() . Let f :
Card be dened by f() = [B

x
[ +. Note that
has a maximal member or
has a conal subset of cardinality . Now choose
n
by induction on n such that
n

, [
n
[ and n = m+1
m

n
. Let
0
be a conal subset of of cardinality , see above why possible. If n = m+1,
let
n
= f() :
m

:
m

m
. Now
n
: n < is
as required.
2.30
{tp26.47}
Discussion 2.31. Note that P
x
in 2.4 is -directed, but in general it is not denable
in M
x
and even not denable in (M
x
)
[cx]
(or M
[Bx+cx]
) even by disjunction of types
as it depends on

d
x
. So we may consider P
x
= P
M,cx,x
= p : p S
(x)
(A c
x
)
and A M has cardinality < ordered as before. Now P
x
is partially ordered
but it is not clear that it is -directed. Moreover (M
x
)
[cx]
is not -saturated, but
is (D
x
, )- sequence homogeneous for suitable D
x
and D
x
is a good diagram (see
e.g. [Sh:3]). So we can consider the families of such Ds, xing (T and) .
But we can dene the order in the -saturated (M
x
)
[Bx]
which is L
,
(
T
)-
equivalent to C
[Bx]
. In this model we have ( x, y) L
,
which is a partial order
on the -tuples, g( x) =
x
= g( y).
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

34 SAHARON SHELAH
However, in our case we know more. Letting = L(
T
), if cf() > [T[ we
know that we can nd

=
( x
, y
, z
) : ) and the order on the set of

e = (. . . e
. . .)
such that in 2.14 we can choose =
(easy, see 2.19(0)). If

cf() > 2
, we can x there also the type of e over C

x

d
x
.
So
()
1
let I = e : e as above, so I is type-denable in (M
x
)
[Bx]
()
2
p
e
=
( x
, e, c
x
) : for e I
()
3
(a)
1
dened by e
1
e
2
if p
e2
p
for some p
such that
p
e1
p
S(B
x
e
1
)
(b) R is dened by e
1
Re
2
i
( x, e
, c) p
( x,
b, c):
b Rang( e
1
B
x
) for each where p
is as above.
There are other variants, we intend to return to this.
We now consider some variants.
{tp.77}
Denition 2.32. 1) In Denition 2.6 we add and dene K
, K
,,
, etc., also for
= 2, 3 by replacing clause (f)
1
by (f)
where
(f)
2
if = 2 then tp( c
t
, A c
s,m
: s <
I
t, m < n
s
(f)
3
if = 3 then for some endless indiscernible sequence

b
t
=
b
t,r
: r J
t
)
of sequences from B, the sequence c
t,0
realizes
13
the type Av(A c
s,m
:
s <
I
t, m < n
t
,

b
t
).
2) We dene mxK
,,
similarly.
{tp.84}
Claim 2.33. 1) K
3
K
2
K
1
.
2) If 2, 3 and x
: < ) is
1
-increasing in K
,,<
and < cf(), < cf()
then x
<
x
dened as in 2.7 belongs to K
,,<
and is a
1
-lub of the
sequence.
3) If = 0, 1, 2, 3 and

d
>
C and M is -saturated, cf() then x = (M, , <>
,

d) K
,<
.
4) Like 2.11 for = 2, 3, i.e.: if cf() > [T[, cf() > , 3 and x K
,,<
then
14
for some y we have x
1
y mxK
,,<
; so in tp(c
x
, M
y
+c
x
) we can get
tp( c, M) is nitely satisable in M
y
+ (C
y
C
x
)).
5) If x K
2
and c C
x
then tp( c, A
x
x
.
{tp.98}
Claim 2.34. 1) In 2.13 we can deal with K
,,<
, = 2, 3, i.e. if = 2 we should
strengthen the assumption to q is nitely satisable in B
.
2) If = 3 we should strengthen the assumption to q = Av(M
x
C
x
, I), I an endless
indiscernible sequence of cardinality < .
3) In 2.14 we can deal with mxK
2
,,
.
13
we may consider ct realizes seeming this makes no dierence
14
so if = cf() > |T|, M is -saturated and

d

+
>
C then for some c

+
>
C and
B [M]
<
we have (M, B, c,

d) mxK
2
.
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

3. Existence of strict type decompositions
We here get a better decomposition, but at a price using normal ultralter (so
measurable or supercompact cardinals).
{pr.7}
Hypothesis 3.1. 0) T is dependent, C = C
T
a monster; we may add one of: (1) -
(4).
(1) (a) = is a measurable cardinal
(b) D is a normal ultralter on I = , so I is a linear order
(c) M
C is -increasing, |M
| <
(d) M =
<
M
, by clauses (c) + (f) necessarily is saturated

(e)
I
(t) := t for t I, this notation is introduced only for having
a uniform treatment of (1) and(2)
(f) M
is |M
|
+
-saturated for <
(2) (a) I is the following partial order, which is (< )-directed:
() set of elements a []
<
: a and
() s
I
t i s t [s[ < min(t)
(b) D is a ne normal ultralter on I and it follows that is a measurable
cardinal
(c) M
t
C, |M
t
| < and s <
I
t M
s
M
t
(d) M = M
t
: t I
(e)
I
(t) = Min(t) for t I
(f) if s <
I
t then M
t
is |M
s
|
+
-saturated
(3) like (2) without the normality
(4) (a)(b), (c), (d) of part (2).
Remark 3.2. So in 3.1 we have (1) (2) (3) and (2) (4).
{pr.14}
Denition 3.3. [under 3.1(1) or (2) or just see (4) so these notions depend on
M
t
: t I).]
1) For U I (usually D), so is a partial order, we say a
t
: t U ) is indiscernible
in M over A when (A M and):
(a) g( a
t
) is constant possibly innite, and a
t
M for t U
(b) for each n for some p
n
for every t
0
<
I
. . . <
I
t
n1
from U we have
tp( a
t0
. . . a
tn1
, A, M) = p
n
.
2) We say a
t
: t U ) is fully indiscernible (in M) over A when (a),(b) above and
(c) if s <
I
t are from U then a
s
M
t
(d) if s U then
15
the sequence a
t
: t U I
s
) is indiscernible over M
s
A
where, of course, I
s
:= t I : s
I
t.
3) In parts (1),(2) of the denition we say k-indiscernible when in clause (b) we
demand n k.
We shall in 3.7 below prove the existence of:
15
by normality (i.e. if (1) or (2) or (4) holds) then this follows
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

36 SAHARON SHELAH
{pr.35}
Denition 3.4. 1) For an innite linear order J we say ( c
t
,

d
t
) : t J) is a strict
(, < )-decomposition over (M, B), B M (and over M means for some B M
of cardinality < ) when :
(a) M C is -saturated, but if we write <
0
instead of we mean M C
and if we write 0 instead of we replace M by a set B
(b) = g( c
t
), = g(
d
t
) are <
(c) if t
0
<
J
. . . <
J
t
k
then tp( c
tn
, M + c
t0
d
t0
+. . . + c
tn1
d
tn1
) is increasing
with n and does not split over B
(d) ( c
t
d
t
: t J) is indiscernible over M
(e) if s <
J
t then tp(
d
t
, c
t
+

d
s
) tp(
d
t
, c
r
d
r
: r
J
s
c
r
: r JM)
(f) for every A M of cardinality < for some c
d
+
M, the sequence
c
d) c
t
d
t
: t J) is an indiscernible sequence over A, so if = 0 this is
an empty demand.
2)
(A) If =
+
instead of < we may write .
(B) If = 0 then M is replaced by a set A, if we write <
0
instead of then
M is just a model.
(C) We say strict
1
(, )-decomposition if (in part (1)) we omit clauses (e) and
(f).
(D) We say strict
0
(, < )-decomposition if we omit (f) and weaken (e) to (e)
,
where
(e)
if s <
I
t then tp(
d
t
, c
t
+

d
s
) tp(
d
t
, c
r
d
r
: r s c
t
M).
(E) Strict may be written strict
1
.
A natural question about those notions of indiscernibility is about existence results.
Now 3.5 is known set-theoretic existence and 3.7 is existence for dependent theories.
{pr.21}
Fact 3.5. [under 3.1(1) or (2) or just (4)]
1) If A M, [A[ < , < , U
1
D and a
t

C for t U
1
then for some U
2
U
1
from D the sequence a
t
: t U
2
) is indiscernible over A.
2) In fact fully indiscernible.
Proof. 1),2) By well known set theory (see Kanamori Magidor [KM78]).
3.5
{pr.23}
Observation 3.6. 1) For some U
D, for every t I the model M

t
is (t)-
saturated and (T) > [T[.
2) If M C is -saturated, J an innite linear order
0
, B [M]
<
and
c
t
= ) =

d
t
for t J then ( c
t
:

d
t
) : t J) is a strict
(, < )-decomposition over

(M, B).
Proof. Obvious.
3.6
{pr.28}
Claim 3.7. Assume 3.1(1) or (2) and > [T[ so the -saturated M C
is well dened and . For every () <
+
and

d
()
C there are B and
( c
t
,

d
t
) : t U ) such that:
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

(a) U D,
(b) B M, [B[ < ,
(c) g(
t
) = () + ,
(d)

d

d
,
(e) x = (M, B, c
,

d
) mxK
,
,
(f) c
t
M realizes tp( c
, M
t
) for t U ,
(g) c
t
: t U ) is fully indiscernible (in M) over B and even over
B c
, (see Denition 3.3(2)),

(h)
1
if t
0
<
I
. . . <
I
t
m
<
I
t
m+1
. . . <
I
t
n
belongs to U then tp(
tm+1
, c
tm+1
+
c
tm+2
+ . . . + c
tn
+

d
tm
) tp(
tm+1
, c
t0
+ . . . + c
tm+1
+ . . . +
c
tn
+

d
t0
+ . . . +

d
tm
+ M
t0
),
(h)
2
if s <
I
t are from U then c
s
is from M
t
, (actually follows
from clause (g))
(h)
3
if t
0
<
I
. . . <
I
t
n
are from U then tp(
t1
,

d
t0
+ c
t1
) tp(
t1
, M
t0
+
t0
+ c
t0
+ c
t1
+ . . . + c
tn
).
{pr.29}
Remark 3.8. We easily can add
(i) x is normal, i.e. Rang( c
) Rang(
).
Proof. First by induction on n we choose

d
n
, c
n
, B
n
, ( c
n
t
,

d
n
t
) : t U
n
) and if n > 0
also e
n
, e
n
t
(for t U
n
) such that
n
(a)

d
n

()+n
C and

d
0
=

d
(b) x
n
= (M, B
n
, c
n
,

d
n
) mxK
,
is normal
(c)

d
m

d
n
if m < n
(d) c
m
= c
n
I
xm
if m < n
(e) c
n
t

d
n
t
is from M and realizes tp( c
n
d
n
, M
t
) for t U
n
(f)

d
n
=

d
m
e
n
and g( e
n
) = and x
m

1
x
n
if n = m + 1
(g) c
n
t

d
n
t
: t U
n
) is fully indiscernible over B
n
+ c
n
+

d
n
(h) U
n
D decrease with n and is U
from 3.5
(i) if s < t are from U
n
then M
s
+ c
n
s
+

d
n
s
M
t
(follows by (g))
(j) tp(
d
m
, e
n
t
+ c
m
) tp(
d
m
, M
t
+ c
m
+ c
m
t
+

d
m
t
) if n = m+ 1, t U
n
(k) (M
t
, B
n
, c
n
,

d
n
) mxK
I(t),
for t U
n
(l) if n = m+ 1, k and t
0
<
I
. . . <
I
t
k
are from U
n
then
tp(
d
m
t1
, c
m
t1
+ e
n
t0
) tp(
d
m
t1
,
k
=0
c
m
t
+

d
m
t0
+ e
n
t0
+ M
t0
).
Case 1: n = 0.
First, let

d
0
=

d. Recalling 2.7(4) clearly y
n
=: (M, , <>,

d
0
) K
,
.
Second, by Claim 2.11 we can nd B
n
, c
n
such that y
n

1
x
n
:= (M, B
n
, c
n
,

d
n
)
mxK
,
.
Third, for t I we can choose c
n
t

d
n
t
from M which realizes tp( c
n
d
n
, M
t
).
Fourth, by 3.5 we choose U
0
D such that c
n
t

d
n
t
: t U
0
) is a fully in-
discernible sequence over B
0
and (by the normality of the lter D) in particular
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

38 SAHARON SHELAH
c
n
s
d
n
s
M
t
when s < t U
0
are from I and U
0
is U
from 3.5. It is easy to

check that all the demands hold.
Case 2: n = m + 1.
First, by clause (k) for m and Conclusion 2.20 for each t U
m
recalling M
t
is
(t)-saturated, (t) > [T[ and M is -saturated there is e
n,
t

M such that
()
1
tp(
d
m
, c
m
+ e
n,
t
) tp(
d
m
, c
m
+ M
t
+ c
m
t
+

d
m
t
).
Second, by 3.5 choose U
n
U
m
which belongs to D such that c
m
t

d
m
t
e
n,
t
: t
U
n
m
+ c
m
+

d
m
.
We shall now prove
1
if t
0
< . . . < t
k
are from U
n
then tp(
d
m
t1
, c
m
t1
+ e
n,
t0
) tp(
d
m
t1
, c
m
t1
+ . . . +
c
m
t
k
+ M
t0
).
By clause (e) of
m
we have
()
2
tp( c
m
t1
d
m
t1
, M
t1
) = tp( c
m
d
m
, M
t1
).
By ()
2
there is an elementary mapping f mapping c
m
d
m
to c
m
t1
d
m
t1
which is the
identity on M
t1
. But M
t0
+ c
m
t0
+

d
m
t0
M
t1
by
m
(i) and e
n,
t0
M
t1
by the full
indiscernibility, i.e. by the choice of U
n
above hence by applying f on ()
1
for t
1
and monotonicity we get
()
3
tp(
d
m
t1
, c
m
t1
+ e
n,
t0
) tp(
d
m
t1
, M
t0
+ c
m
t1
+ c
m
t0
+

d
m
t0
).
Now by clause (k) of
m
()
4
(M
t0
, B
m
, c
m
,

d
m
) mxK
I(t0),
.
But by clause (e) of
m
()
5
tp( c
m
t1
d
m
t1
, M
t0
) = tp( c
m
d
m
, M
t0
).
By ()
4
+ ()
5
()
6
(M
t0
, B
m
, c
m
t1
,

d
m
t1
) mxK
I(t0),
.
Also easily by clauses (k) + (i) of
m
applied to t = t
2
, t
3
, . . . , t
k
(recalling 1.4(3))
()
7
tp( c
m
t2
. . . c
m
t
k
, M
t0
+ c
m
t1
+

d
m
t1
m
.
By ()
6
+ ()
7
and the weak orthogonality claim 2.13(1) we have
()
8
tp(
d
m
t1
, M
t0
+ c
m
t1
) tp(
d
m
t1
, M
t0
+ c
m
t1
+ . . . + c
m
t
k
).
By ()
3
+ ()
8
()
9
tp(
d
m
t1
, c
m
t1
+ e
n,
t0
) tp(d
m
t1
, M
t0
+ c
n
t1
+ . . . + c
m
t
k
)
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

as promised in
1
.
Now we continue to deal with Case 2, choose F
n
: U
n
U
n
such that s
U
n
s <
I
F
n
(s) U
n
and for t U
n
we let e
n
t
:= e
n,
Fn(t)
. Let e
n

C be
such that c
m
d
m
e
n
realizes Av( c
m
t

d
m
t
e
n
t
: t U
n
)/D, M). Let U
n
D be
U
n
and such that s U
n
t U
n
s <
I
t F
n
(s) <
I
t and the sequence
c
m
t

d
m
t
e
n,
F(t)
: t U
n
) is fully indiscernible over c
n
d
m
e
n
.
Let

d
n
=

d
m
e
n
and

d
n
t
=

d
m
t
e
n
t
for t U
n
.
Let y
n
:= (M, B
m
, c
m
,

d
n
) K
,
so clearly x
m

1
y
n
hence by 2.11 there
is x
n
= (M, B
n
, c
n
,

d
n
) mxK
,
such that y
n

2
x
n
so c
n
and B
n
are well
dened
16
and c
m
= c
n
I
xm
. For t U
n
, let c
n
t
be a sequence from M such
that c
m
t
= c
n
t
Dom( c
m
) and tp( c
n
t

d
n
t
, M
t
) = tp( c
n
d
n
, M
t
), this is possible as
M is -saturated, [M
t
+ c
m
t
+

d
n
t
[ < , [Rang( c
n
)[ < and c
m
t

d
n
t
realizes the type
tp( c
m
d
n
, M
t
).
Lastly, let U
n
be a subset of U
n
which belongs to D such that
2
c
n
t

d
n
t
: t U
n
n
+ c
n
+

d
n
3
(M
t
, B
n
, c
n
,

d
n
) mxK
,
for every t U
n
.
[Why U
n
exists? By 3.5 and 2.10.]
It is easy to check that x
n
, c
n
,

d
n
,

d
n
, (c
n
t
,

d
n
t
, e
n
t
) : t U
n
) are as required.
E.g. clause (f) holds as g(
d
0
) = g(
d) = () and g(
d
m
) = () + m by clause
(a) of
m
and g( e
n
) = by
n
(f) we can prove that g(
d
n
) = g(
d
m
) + =
() + m+ = () + n, so we clearly are done.
So we can carry the induction. Let c
= c
n<
c
n
,

d
=

d
t
=
n<
d
n
t
and
B = B
n
: n < and U = U
n
: n < .
Let us check that holds indeed.
Clause (a): U D as each U
n
D by
n
(h) and D is -complete and >
0
recalling U = U
n
: n < .
Clause (b): B [M]
<
as B
n
M, [B
n
[ < by
n
(b) for n < and is regular
uncountable recalling and B := B
n
: n < .
Clause (c): By
n
(a) + (c) for n < .
Clause (d):

d =

d
0
=

d
() is proved as in clause (c).

Clause (e): As x
n
= (M, B
n
, c
n
,

d
n
) mxK
,
by clause (b) of
n
and x
n

1
x
n+1
by
n+1
(f) and (M, B, c
,

d
) = x
n
: n < , clearly by claim 2.13(2) we are
done.
Clause (f): By clause (e) of
n
(and the choice of c
,

d
, c
t
,

d
t
, etc).
Clause (g): Similarly by clause (g) of
n
.
Clause (h)
1
: By clauses (h)
2
+ (h)
3
proved below.
Clause (h)
2
: By clause (i) of .
16
in fact, we can demand that tp(cn (Ixn
\Ixm
), M + cm) does not split over Bxn
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

40 SAHARON SHELAH
Clause (h)
3
: Holds by clause () of
n
.
3.7
{pr.42}
Claim 3.9. 1) If = |M|,

d

+
>
C and the conclusion of Claim 3.7 holds (e.g.
if Hypothesis 3.1(1) or (2) holds) then we can nd a strict (, )-decomposition
( c
n
,

d
n
) : n < ) over M such that

d

d
0
.
2) Instead Hypothesis 3.1 it is enough to demand: if M is -saturated and p =
p(x
i
: i < )) is a type with parameters from M C, [C[ which is (< )-
satisable in M, i.e. every subset of p of cardinality < is realized in M then p
can be extended to p
+
S
(M C) which is (< )-satisable in M.

Proof. 1) Easy (as assuming g(
d) = or g() <
+
does not matter).
2) The idea is to repeat the proof of 3.7, but as of unclear value we leave it to the
reader.
3.9
{pr.56}
Claim 3.10. Assume = > [T[ is weakly compact, M
EC
<
(T) is -
increasing continuous, M = M
: < is saturated. Then 3.7 and 3.11(2)

hold.
Proof. Revise the proof of 3.7, but in
n
weaken clauses (g),(h) to (g)
,(h)
and
use the proof of 4.12 in the end where
(g)
U
n
[]
decreasing with n and c

n
t

d
n
t
: t U
n
) is just fully n-indiscernible
(h)
U
n
does not belong to the weakly compact ideal.
We leave the details to the reader.
3.10
{pr.49}
Claim 3.11. 1) (do not depend on 3.1)
Assume M is -saturated, [T[ < , tp( c
n
, M + c
0
+. . . + c
n1
) does not split
over B where [B[ < , B M and g( c
n
) <
+
and

d
n
= c
n
u for n < (so
u Dom( c
n
)) then ( c
n
,

d
n
) : n < ) is a strict (, )-decomposition over (M, B).
2) (Relying on 3.1!) Assume ( c
n
,

d
n
) : n < ) is a strict (, )-decomposition over
(M, B). For any

d

+
>
C we can nd a strict (, )-decomposition ( c
+
n
,

d
+
n
) : n <
) over M such that c
n
c
+
n
,

d
n

d
+
n
for n < and

d
0
d

d
+
0
.
Remark 3.12. We can add (relying on 3.1):
If
d
n
: n < ) is an indiscernible sequence over M (so

d
n

+
>
C) such that for
any A [M]
<
some

d
M realizes tp(
d
0
, A
d
1+n
: n < ) then we can nd
c
n

+
>
C for n < such that (
d
n
, c
n
) : n < ) is a strict (, )-decomposition.
Proof. 1) Easy.
2) Repeat the proof of 3.7 starting with x
0
= (M, B, c
0
,

d
0
) and c
t
d
t
: t I) such
that c
t
d
t

g( ct)+g(
dt)
M realizing tp( c
0
d
0
, M
t
+
n<
c
1+n
d
1+n
).
3.11
{pr.84}
Claim 3.13. The sequence ( c
2
t
,

d
2
t
) : t I
2
) is a strict
0
(0, )-decomposition over
(B
2
, B
1
) when
(a) ( c
1
s
,

d
1
s
) : s I
1
)) is a strict
0
(0, )-decomposition (and g( c
1
s
) = g( c
2
t
), g(
d
1
s
) =
g(
d
1
s
) = g(
d
2
t
), of course)
(b) for any n if I
[= t
,0
< . . . < t
,n1
for = 1, 2 then tp( c
1
t1,0
d
1
t1,0
. . .
c
1
t1,n1
d
t1,n1
, B
2
) = tp( c
2
t2,0
d
2
t2,0
. . . c
2
t2,n1
d
2
t2,n1
, B
2
).
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

Proof. Should be clear.
3.13
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

42 SAHARON SHELAH
4. Consequences of strict decomposition
Here we look again at the generic pair conjecture ([Sh:877, 0.2]). The non-
structure side (in a strong version) is proved there for =
<
non-strong limit
and in [Sh:906] for =
<
strong limit (i.e. strongly inaccessible).
The conjecture is (the instances of G.C.H. are used to make the conjecture trans-
parent):
{ps.1}
Conjecture 4.1. The generic pair conjecture
Assume
17
=
<
> [T[, 2
=
+
, M
EC
(T) is -increasing continuous

for <
+
with M
: <
+
EC
+(T) being saturated. Then T is

dependent i for some club E of
+
for all pairs < <
+
from E of conality
+
, (M
, M
) has the same isomorphism type.

Here we prove the structure side when is measurable. It seemed natural to
assume that the rst order theories of such pair is complicated if T is independent
and understandable for dependent of T.
In fact, it may be better to ask
{ps.3}
Problem 4.2. Assume [T[ < =
<
<
2
=
<2
and M
1
M
2

C, M
1
is
1
-saturated of cardinality , M
2
is
2
-saturated of cardinality . What
can we say on Th(M
2
, M
1
)? On Th
L
,
((T))
(M
2
, M
1
)?
More generally
{gs.3}
Problem 4.3. Assume n < , [T[ < , < <
0
,
=
<
for n,
<
+1
for < n. Let M
be
-saturated of cardinality
for n and M
M
+1
for < n. What can we say on M
+
= Th(M
n
, . . . , M
1
, M
0
),i.e. M
n
expanded
by unary predicates for M
for < n? When can we interpret (with rst order

formulas with parameters) second order logic on ? i.e. classify T by this. Similarly
for L
,
(
M
+).
The proof here, if e.g. = is measurable say that even the L
,
-theory is
constant, but does not say much even on the rst order theory, (see [KpSh:946]).
It is known that, e.g. for T = Peano arithmetic, in Th(M
2
, M
1
) we can interpret
second order logic on (, =).
On n-independent theories see [Sh:886, 2]. Note that
{gs.4}
Claim 4.4. Assume T is the model completion of T
0
, dened below so seems the
simplest 2-independent theory. If M
1
EC
(T) and M
1
is a
+
-saturated -
extension of M
0
then in (M
1
, M
0
) we can interpret second order logic on M
0
(i.e.
quantication on two-place relations, when
()
T0
consists of P
0
, P
1
, P
2
(unary predicates) and R (a ternary predicate)
and a
T0
-model M is a model of T
0
i P
M
0
, P
M
1
, P
M
2
) is a partition of
[M[ and R
M
P
M
0
P
M
1
P
M
2
.
Proof. Obvious.
4.4
Also for T = theory of Boolean algebras (which is n-independent for every n) the
theory is complicated. Of course, it would be better to eliminate the measurable
assumption.
17
the 2
=
+
is just for making the formulation more transparent, and by absoluteness is
equivalent
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

We rst connect decomposition (i.e. the results of 3.7) and sucient conditions for
being an indiscernible sequence.
{pu.14}
Claim 4.5. Assume (1), (2) are xed ordinals
(a) c

(1)
C,

d

(2)
C for <
, let a
= c
(b) tp( a
, B) = p for < <
(c) if < <
then tp(
, c
+

d
+ B) tp(
, c
: B)
(d) tp( c
, a
: < B) is increasing with

(e) tp( c
, a
: < B) does not split over B.

Then a
: <
) is an indiscernible sequence over B.

Proof. For u, v
let B
u,v
=
: u
: v B. For u, v
and increasing function h from u v to
let f = f
hu,hv
be dened as follows:
(a) Dom(f
hu,hv
) = B
u,v
(b) f B = id
B
(c) f maps a
to a
h()
for u
(d) f maps c
to c
h()
for v.
Is f
hu,hv
a well dened function and even one to one? For this it suces to check
the following three demands, which follows by clause (b) of the assumption
()
1
() if , <
, b B and i < g( a
) then ( a
)
i
= b ( a
)
i
= b
() if , <
and i, j < g( a
) then ( a
)
i
= ( a
)
j
( a
)
i
= ( a
)
j
() if
1
<
2
<
and
1
<
2
<
and i, j < g( a
) then
( a
1
)
i
= ( a
2
)
j
( a
1
)
i
= ( a
2
)
j
.
We prove by induction on n that a
: <
) is an n-indiscernible sequence over

B (if n <
). For n 2 this is trivial by clause (b) of the assumption. So assume

n = m + 1 > 2 and we have proved it up to m. So let
0
< . . . <
m
<
,
0
<
. . . <
m
<
and we shall prove that a

0
. . . a
m
, a
0
. . . a
m
realize the same
type over B, this suces.
Now by symmetry without loss of generality
m

m
let h
0
=
) : <
m, h
1
= (
) : < m, h
2
= h
1

m
,
m
) and h
3
= h
0

m
,
m
) and
h
4
= h
0

m
,
m
).
Let f
0
be the mapping f
h0,h0
. By the induction hypothesis
()
2
f
0
Let f
1
be the mapping f
h1,h2
, now by clause (d) of the assumption
()
3
f
1
By ()
2
we know that a
0
. . . a
m1
and f
0
( a
0
. . . a
m1
) realize the same
type over B and they are included in B
m,m
, B
m,m
respectively. But
m

m
so both sequences are from B
m
hence by clause (e) of the assumption, i.e. as
tp( c
m
, B
m,m
) does not split over B, recalling ()
2
we have
()
4
f
2
:= f
h0,h3
By ()
3
+ ()
4
, comparing we have
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

44 SAHARON SHELAH
()
5
f
3
:= f
h0,h4
being f
2
f
1
Note that
()
6
f
3
( c
m
d
m1
) = c
m
d
m1
.
By clause (b) of the assumption and ()
5
and ()
6
clearly
()
7
f
3
(tp(
d
m
, c
m
+

d
m1
+ B)) = tp(
d
m
, c
m
+

d
m1
+ B).
By clause (c) of the assumption
()
8
tp(
d
m
, c
m
+

d
m1
+ B) tp(
d
m
, c
m
+ a
0
+ . . . + a
m1
+ B)
()
9
tp(
d
m
, c
m
+

d
m1
+ B) tp(
d
n
, c
n
+ a
0
+ . . . + a
m1
+ B).
Together f
4
= f
h4,h4
is an elementary mapping and it maps a
to a
for m
(and extend id
B
) so we are done.
4.5
{pu.16}
Observation 4.6. Assume
(a) n <
(b) the linear order I is I
0
+ . . . + I
n
where I
is innite for = 1, . . . , n 1
(c) ( c
t
,

d
t
) : t I
) is a strict
0
(0, )-decomposition over (B
, B
) for n
(d) c
t
d
t
: t I
+ I
+1
) is indiscernible over B
for < n
(e) B
B B
0
(f) B
+1
c
t
d
t
: t I
.
Then ( c
t
,

d
t
) : t I) is a strict
0
(0, )-decomposition over (B
0
, B).
Proof. Straightforward: we prove by induction on k n that c
t
d
t
: t I
:
[n k, n]) is an indiscernible sequence over B
nk
. For k = 0, 1 this holds
by clause (d), for k 2, let s
0
<
I
. . . <
I
s
m1
, t
0
<
I
. . . <
I
t
m1
be from
I
: [n k, n] be given. Choose s
i
, t
i
< m from I
nk
I
nk+1
for i < m
such that s
0
<
I
. . . <
I
s
m1
, t
0
<
I
. . . <
I
t
m1
and s
I
nk
s
= s
, s
/
I
nk
s
I
nk+1
, t
I
nk
t
= t
, t
/ I
nk
t
I
nk+1
. This is possible
as I
nk+1
is innite and k [2, n].
Let k m e such that < m (s
I
nk
< k); so by the induction hy-
pothesis we have tp( a
s
k
. . . a
sm1
, B
nk+1
) = tp( a
s
k
. . . a
s
m1
, B
nl+1
). As <
k a
s
= a
s
B
nk+1
and B
nk
B
nk+1
it follows that tp( a
s0
. . . a
sm1
, B
nk
)
is equal to tp( a
s
0
. . . a
s
m1
, B
nk
). This type by clause (d) is equal to tp( a
t
0
. . . a
t
m1
, B
nk
).
Similarly to the proof in the beginning of the paragraph, this type by the induction
hypothesis is equal to tp( a
ti
. . . a
tm1
, B
nk
), so together we are done.
4.6
The following is a local version of 4.5
{du.19}
Claim 4.7. Assume (n() < and () > n())
(a) c
m

(m,1)
C for m n(), < ()
(b)

d
=

d
0

(m,0)
C for < ()
(c) c
m
c
m+1
for < (), m < n()

(d) e
m

(m,2)
C for non-zero m n(), < ()
(e) for each k < n(), for all < < () the type tp( c
k+1
e
k+1
e
k
, B)
is the same
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

(f) tp(
e
m
, c
m
+ e
m+1
+ B) tp(
d
m
e
m
, c
m
<
c
m
+ B)
(g) tp( c
k
, c
k+1
e
k+1
: < , m < n() B) is increasing with

(h) tp( c
k
, c
k+1
e
k+1
: < and m < n() B) does not split over B

(i) tp( c
n()
e
k
, B) is the same for all < ().

Then c
0
: <
) is an n()-indiscernible sequence over B.

Remark 4.8. 1) In what sense is 4.7 a local version of 4.5? In the second we get only
n()-indiscernibility. Note that the role of

d
there is played by

d
, e
m
(m < n())
here.
2) The claim is not used in the rest of the section.
Proof. We prove by induction on n < n() that
1
if k n() n and
0
< . . . <
n
and
0
<
1
< . . . <
n
then
c
k
0
d
0
. . . c
k
n1
d
n1
c
k
n
d
n
e
k
n
and
c
k
0
d
0
. . . c
k
n1
d
n1
c
k
n
d
n
e
k
n
For n = n() 1, k = 0 we get the desired conclusion.
For n = 0 this holds by clause (i) of the assumption. So assume n = m + 1 and
k n() n and we have proved this for m. Note that k + 1 n() m. So
let
0
< . . . <
m+1
<
,
0
< . . . <
m+1
<
and without loss of generality
n

n
and we shall proof the equality of types from
1
in this case, this suce.
Now
()
1
c
k+1
0

d
0
. . . c
k+1
m1
d
m1
c
k+1
m

d
m
e
k+1
m
and
c
k+1
0
d
0
. . . c
k+1
m1
d
k+1
m1
c
k+1
m

d
m
e
k+1
m
realize
the same type over B.
[Why? By the induction hypothesis.]
()
2
tp( c
k
n
, c
k+1
e
k+1
: <
n
B) extends tp( c
k
n
, c
k+1
e
k+1
:
<
n
B) and does not split over B.
[Why? By clause (g) and by clause (h) of the assumption.]
()
3
tp( c
k
n
, c
k+1
e
k+1
: <
n
B) does not split over B.
[Why? By clause (h) of the assumption.]
()
4
c
k+1
0

d
0
. . . c
k+1
m1
d
m1
c
k+1
m

d
m
e
k+1
m
c
k
m+1
and
d
k+1
0
d
0
. . . c
k+1
m1
d
m1
c
k+1
m
d
m
e
k+1
m
c
k
m+1
realize
the same type over B.
[Why? By ()
1
+ ()
3
.]
()
5
in ()
4
we can replace c
k
m+1
in the rst sequence by c
k
m+1
.
[Why? By ()
4
+ ()
2
.] But
()
6
( c
k+1
m+1
e
k+1
m
)(
d
m+1
e
k
m+1
) and ( c
k+1
m+1
e
k+1
m
)(
d
m+1
e
k
m+1
) realize the
same type over B.
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

46 SAHARON SHELAH
[Why? By clause (e) of the assumption.]
()
7
c
k+1
0

d
0
. . . c
k+1
m1
d
m1
c
k+1
m

d
m
c
k+1
m+1
d
m
m+1
e
k
m+1
and
c
k+1
0
d
0
. . . c
k+1
m1
d
m1
c
k+1
m
d
m
c
k+1
m+1
d
m+1
e
k
m+1
[Why? By ()
5
+ ()
6
and clause (f) of the assumption.]
As c
k
c
k+1
by clause (c) of the assumption, we nish the induction step. Hence

we get the desired statement.
4.7

We now return to the generic pair conjecture. Central here is the following deni-
tion; the best case is = = is a measurable cardinal.
{ps.7}
Denition 4.9. We say that the triple (, , < ) is good or T-good when :
(A) [T[ < = cf() = cf() =
<
(B) T is dependent
(C) if M is -saturated of cardinality then
(a) if

d
>
C then we can nd B and a strict (, < )-decomposition
( c
n
,

d
n
) : n < ) over (M, B), see Denition 3.4, such that

d

d
0
(b) moreover 3.11(2) holds (true if 3.1 holds).
So to begin our analysis we need
{ps.11}
Claim 4.10. If T is a dependent and > [T[ is a measurable cardinal then (, , <
) is T-good.
Proof. Immediate by 3.11(2) and 3.6(2), you may use 3.7, too.
4.10
For the rest of this section we assume
{ps.13}
Hypothesis 4.11. 1) T is dependent.
2) [T[ < and =
<
> [T[ and (, , < ) is T-good.
{ps.21}
Claim 4.12. Assume
(a) is a limit ordinal < = cf()
(b) ( c
n
,

d
n
) : n < ) is a strict (, < )-decomposition over (M, B
) for each
< recalling Denition 3.4
(c) c
n
c
n

d
n

d
n
for < < , n <
(d) B
for < <

(e) we dene c
n
= c
n
: < ,

d
n
=
n
: < , B
= B
: < .
Then ( c
n
,

d
n
) : n < ) is a strict (, < )-decomposition over (M, B
).
Proof. We have to check clauses (a)-(f) of Denition 3.4. Clause (a) is trivial by
assumption (b) of 4.12, clause (b) holds as < = cf() by assumption (a) of 4.12
and clauses (c),(d),(e) hold by their local character and assumptions (b) + (c) of
4.12. To prove clause (f) is the main point, it means to show
1
if B
A M and [A[ < then for some pair ( c,

d) of sequences from M
we have c
d) c
n
: n < ) is an indiscernible sequence over A.
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

Now by induction on < we choose a pair ( c
,

d
) such that
1
c
)c
n
: n < ) is an indiscernible sequence over A
:
< .
[Why possible? We can choose ( c
,

d
) because ( c
n
,

d
n
) : n < ) being a strict
(, < )-decomposition over B
, we can apply clause (f) of Denition 3.4 recalling

B
being a subset of A( c
: < and the later being M and of cardinality

< as is regular > .]
We can nd ( c
,

d
) : < ) though not necessarily in M such that
2
(a) ( c
) : < ) c
n
: n < ) is an indiscernible sequence over A
(b) c
and

d
for < .
[Why? For this by using the saturation of C, it is enough to prove that: if
n < m < ,
0
< . . . <
n1
< then the sequences c
0
. . . c
n1
n1
c
0
0

d
0
0
c
1
1

d
1
1
. . . c
n1
n1

d
n1
n1
realize the same type over A c
n
k
: k [n, m).
This is proved by induction on n < and is straight.]
Hence we can nd a pair ( c
,

d
) such that
3
c
)( c
) : < ) c
n
:

d
n
: n < ) is an indiscernible sequence over
A.
Lastly, we choose ( c
,

d
) such that
4
( c
,

d
) is a pair of sequences from M realizing tp( c
, A
: <
, of course with g( c
) = g( c
0
), g(
) = g(
0
) equivalently there is an
automorphism of C which is the identity on A
: < mapping
c
to c
.
We shall prove that ( c
,

d
) is as required. Now to prove that ( c
,

d
) is as required
in
1
it suces to prove, for each < that
2
(( c
g( c
0
))((
g(
0
)))( c
n
) : n < )
is an indiscernible sequence over A.
[Why? As
1
is a local demand, i.e. it says that c
is a sequence realizing
an appropriate type q (and is from M) and for this it suces to check every nite
subtype so
2
suces.]
Now
2
follows by
6
below. Let ( c
,
,

d
,
) = ( c
g( c
),

d
g(d
0
)) and
( c
,

d
)) = ( c
g( c
0
),

d
g(
0
)) for < and [, ).
Now
5
( c
,

d
))( c
,
,

d
,
) : [, )) is a strict
0
decomposition over (A, A).
[Why? By
3
, this holds for ( c
g( c
0
),

d
g( c
0
)))( c
,
,

d
,
) : [, ))
and we use preservation by automorphism of C, i.e. use
4
.]
6
For < and i the sequence ( c
,

d
))( c
,
,

d
,
) : [, ))(c
n
,

d
n
) :
n < i) is a strict
0
decomposition over (A, A).
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

48 SAHARON SHELAH
[Why? We prove this by induction on i, so for i = we get the desired conclusion;
also for i = the inductive step is trivial and for i = 0 use
5
. So assume i = n+1,
let
A
1
= A Rang( c
) Rang(
)
A
2
= A
2,
where
for (, ] we let A
2,
= Rang( c
,
) Rang(
,
) : [, )
A
3
= c
: < n.
Clearly
() B
1
, B
2,
, B
3
are M.
Let (, ) be a successor ordinal and we shall prove that (c
,
) . . . ( c
+n,
+n,
)
and ( c
0
) . . . ( c
n
) realize the same type over A
1
A
2,
, as is a limit ordinal
this suces. Now ( c
,
) . . . ( c
+n1,
+n1,
) and ( c
0
) . . . ( c
n1
n1
)
realize the same type over A
1
A
2,
by the induction hypothesis.
Next tp( c
n
, M
m
: m < n) does not split over A hence tp( c
n
, AA
1
A
2
A
3
A by ().
Hence c
,

d
)) c
,
, d
,
) : [, ))( c
k
,

d
k
) : k < n) is an indiscernible
sequence over A c
n
and ( c
,
,

d
,
) : [, ))( c
k
,

d
k
) : k < ) is an indis-
cernible sequence over A.
Together
(c
d
,
) . . . ( c
+n1
+n1,
) c
+n,
and ( c
0
) . . . ( c
n1
n1
) c
n
realizes the same type over A
1
A
2,
.
By
5
we can choose

d
in C such that ( c
d
,
) . . . ( c
+n1,
+n1,
)( c
+n,
)
and ( c
0
) . . . ( c
n1
n1
)( c
n
) realize the same type over A
1
A
2,
so it
suces to prove that

d
+n,
,

d
realize the same type over A

1
A
2,+n
c
+n,
.
Recall that c
,
: [, )) c
n
: n < ) is an indiscernible sequence,
hence the sequences c
1,
1,
c
,
. . . c
+n,
+n,
and
c
1,
1,
c
d
1
0
. . . c
n
realizes the same type.
By the two previous sentences and the transitivity of the equality of types
+n,
,

d
realize the same type over ( c
1,
1,
)( c
,
) . . . ( c
+n1,
+n1,
) c
+n,
,
but tp(
+n,
, c
+n,
+

d
+n1,
) tp(
+n,
, A
1
+ A
2,+n
+ c
+n,
) so we are
done.]
4.12
{ps.23}
Claim 4.13. If M is saturated and B [M]
<
then the number of strict (, , < )-
decompositions ( c
n
,

d
n
) : n < ) up to equivalence over B is , see below.
{ps.25}
Denition 4.14. 1) We say that the strict (, < )-decompositions ( c
,

d
) :
< ), ( c
,

d
) : < ) are equivalent over B [M]

<
when for some auto-
morphism f of M over B for every n and
0
< . . . <
n1
< , f maps the type
tp( c
) . . . ( c
n1
n
1
), M) to the type tp(( c
0
) . . . ( c
n1
n1
), M)
and g( c
0
) = g( c
0
), g(
0
) = g(
0
).
2) If B = then we may omit it.
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

Proof. Proof of 4.13:
By 1.2(4), that is [Sh:783, 5.26], however in the real case is measurable hence
strongly inaccessible so 4.13 is easy. That is, xing A, B, also the number of <
and p( x
) S(M) not splitting over B is by 1.2.

4.13
Clearly 4.12 is a step forward. Now we prove, using a measurable, the generic
pair conjecture; instead of assuming that the cardinality is measurable we can
restrict T.
Toward this
{ps.28}
Denition 4.15. We say that the triple m = (M, N, A) is a (, , < )-system
when ( = cf() > [T[ and) it satises clauses (a)-(d) below, and say the
triple is a full (, , < )-system when it satises clauses (a)-(h) below where:
(a) M is -saturated of cardinality
(b) M N C and N has cardinality
(c) A is a set of cardinality of objects p such that:
() p is of the form ( c
,

d
) : <
+
) = ( c
[p],

d
[p]) : <
+
)
() c
0
d
0
N
() ( c
,

d
) : <
+
) is a strict (, < )-decomposition over M
(d) A is partially ordered by: p
m
q i
c
0
[p] c
0
[q] and

d
0
[p]

d
0
[q]
for every large enough <
+
, c
[p] c
[q],

d
[p]

d
[q]
(e) A is closed under union of < increasing chains of length <
(f) if p A and

d
>
N then for some q A above p we have
Rang(
d) Rang(
d
0
[q])
(g) N is saturated
(h) up to equivalence every possible p in clause (0) occurs (i.e. is
represented in A)
(i) there is q
2
A such that p
2

m
q
2
witnessed by f and q
2
, q
1
are
equivalent and ( c
0
[q
2
],

d
0
[q
2
]) = ( c
1
,

d
2
) when
() p
1

m
q
1
so both are from A
m
() p
2
A is equivalent to p
1
() ( c,

d) is from N even f aut(M) maps p
1
to p
2
() g( c
2
) = g( c
[q
1
]]), g(
d
2
) = g(
[q
1
])
() f maps the type of c
d c
0
[p
2
]
d
0
[p
2
] over M to the type of
c
0
[q
1
]
d
0
[q
1
] c
0
[p
1
]
d
0
[q
1
] over M.
{ps.35}
Denition 4.16. 1) Let BP = BP
,,<
be the set of (, , < )-systems.
2) If (M
, N
, A
) is a (, , < )-system for = 1, 2 we say (M

2
, N
2
, A
2
) is above
(M
1
, N
1
, A
1
) or (M
1
, N
1
, A
1
)
BP
,,<
(M
2
, N
2
, A
2
) when M
1
= M
2
, N
1
N
2
and A
1
A
2
.
3) We may write just BP,
BP
when (, , < ) is clear from the context.
{ps.42}
Claim 4.17. Assume = =
<
and (, , < ) is T-good (see Denition 4.9).
1) If M C is -saturated of cardinality then there is a pair (N, A) such that
(M, N, A) of a (, , < )-system.
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

50 SAHARON SHELAH
2) If (M, N
1
, A
1
) is a (, , < )-system and N
2
C and |N
2
| = then there is a
pair (N
3
, A
3
) such that (M, N
3
, A
3
) is a (, , < )-system above (M, N
1
, A
1
) and
N
2
N
3
.
3) If (M, N
, A
) : < ) is an increasing sequence of (, , < )-systems and is

a limit ordinal <
+
then the union, (M,
<
N
<
A
) is a (, , < )-system
which is a list upper bound of (M, N
, A
) : < .
4) If in part (3) we have cf() = and each is full, then so is the union.
Proof. 1) Let m = (M, M, ) and check.
2) Let N
3
C be such that N
1
N
2
N
3
, N
3
is saturated of cardinality and use
(M, N
3
, A
1
).
3) Easy.
4) Easy.
4.17
{ps.43}
Claim 4.18. 1) We have:
(a)
BP
is a partial order on BP
(b) any
BP
-increasing sequence of length <
+
has an upper bound
(b)
+
moreover if (N
, A
) : < ) is
BP
-increasing then (N
: <
, A
: < ) is a least
BP
-upper bound
(c) BP is not empty, moreover for every M EC
,
() there is m BP such
that M
m
= M.
2) For m is a (, , < )-system then :
(a)
m
is a partial order of A
m
(b) if also n is a (, , < )-system and m
BP
n then
m
=
n
A
m
(c) if as a limit ordinal < and p
A
m
for < is
m
-increasing
with then there is n BP such that m
BP
n and
n
-upper bound of
p
: <
(d) if p A
m
and

d
>
N then there are a (, , < )-system n satisfying
m
BP
n and q A
n
such that p
n
q and

d Rang(
d
q
0
)
(e) if p
1
, p
2
A
m
are equivalent and p
1

m
q
2
then for some n BP we
have m
BP
n and for some q
2
A
n
equivalent to q
1
we have p
1

n
q
2
.
3) If (M, N, A) is a (, , < )-system then there is a full (, , < )-system above
it.
Proof. 1) Easy, e.g. (c) by 4.17, and for clause (c) use (M, M, ).
2) Why? Easy, we mainly elaborate clause (c). For < < let = (, ) <
+
be such that
if [,
+
) then c
p
c
p
,

d
p

d
p
.
Let () = sup(, ) + 2 : < < , so necessarily () <
+
. Clearly if
[(),
+
) then ( c
p
: < ) is -increasing,
d
p
: < ) is -increasing.
We choose c
0
= c
p
0
: < ,

d
0
=
d
p
0
: < and for <
+
let
c
1+
= c
p
()+
: < and

d
1+
=
d
p
()+
: < .
Now easily
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

p = ( c
,

d
) : <
+
) is well dened and satises the demand in ()
1
(c)()+
().
[Why? By Claim 4.12.]
So we dene n as (M, N
m
, p A
m
), easily
n BP and m n.
Lastly, < p

n
p as witnessed by () . So we are done proving clause
(c).
Why clause (d) holds? Easy by 3.11(2), i.e. by the assumption (, , < ) is
T-good of Claim 4.17.
3) We x a (, , < )-system (M, N
, A
). We now shall choose m

i
BP,
BP
-
increasing by induction on i and
1
for i = 0 : m
0
is (M, N
, A
2
for limit i < , m
i
is a
BP
-upper bound of m
j
: j < i), in fact the union.
[Why? Possible by clause (b)
+
of part (1).]
By bookkeeping
3
if p
: < ) is
mj
-increasing, a limit ordinal < , then for some
i (j, ) the sequence has an upper bound by
mi+1
.
[Why? By clause (c) of part (2).]
4
if p A
mj
and

d
>
N
mj
then for some i (j, ), there is q A
mi+1
such that p
mi+1
q and Rang(
d) Rang(
d
q
0
).
[Why? By clause (d) of part (2).]
5
similarly to
4
for clause (i) of Denition 4.15.
Now m
is as required.
4.18
{ps.44}
Claim 4.19. Assume (, , ) = (, , ) that is (, , < ) is T-good.
If m
= (M
, N
, A
) is a full (, , < )-system for = 1, 2 then (N

1
, M
1
)

=
(N
2
, M
2
) that is there is an isomorphism f from N
1
onto N
2
mapping M
1
onto
M
2
.
Proof. We dene the set AP of approximation:
()
1
AP is the set of triples h = (p
1
, B
1
, p
2
, B
2
, f) = (p
1
[h], B
2
[h], p
2
[h], B
2
[h], f[h])
(a) p
for = 1, 2
(b) B
[M
]
<
is a base for p
(c) f is an elementary mapping which maps B

1
onto B
2
(d) if f
+
is an isomorphism from M
1
onto M
2
extending f and
0
< . . . <
n1
<
+
then f
+
maps tp(( c
p1,1
d
p1,1
) . . . ( c
p1,n1
d
p1,n1
), M
1
)
onto tp(( c
p2,0
d
p2,0
) . . . ( c
p2,n1
d
p2,n1
), M
2
).
()
2
we dene the two-place relation
AP
by h
1

AP
h
2
i
(a) both are in AP
(b) p
[h
1
]
m
[h
2
]
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

52 SAHARON SHELAH
(c) B
[h
1
] B
[h
2
]
(d) f
1
[h
1
] f[h
2
].
Obviously
()
3

AP
partially ordered AP.
Also
()
4
if < is a limit ordinal and h
: < ) is
AP
-increasing, then this
sequence has an
AP
-upper bound.
[Why? As in the proof of 4.18(2)(c).]
()
5
if h AP, 1, 2 and a M
1
then there is h
AP such that h
AP
h
1
and a B
p
[h
1
]
.
[Why? Let a
1
= a and a
2
M
2
realize f
h
(tp(a
1
, B
p1[h]
)).
Let B
= (B
p
[h]
a
) and let p
A
m
be dened by: ( c
[p
],

d
[p
]) =
( c
[p
[h
1
],

d
[p
[h]).
Lastly, let h
= (p
1
, B
1
, p
2
, B
2
, f
h
(a
1
, a
2
)).]
()
6
If h AP, 1, 2 and a M
2
then there is h
AP such that h
AP
h
and a B
2
[h
].
[Why? Like ()
5
.]
()
7
if h AP and d N
2
then for some h
AP we have h
AP
h
and
d Rang(
d
0
[p
1
(h
)]).
[Why? There is q
1
A
m1
such that p
1
[h]
m1
q
1
and d

d
0
[p
1
(h
).]
Let f
f be an isomorphism from M
1
onto M
2
.
Now by clause (i) by Denition 4.15 there is q
2
A
m2
such that f
maps q
1
to
q
2
.
The rest should be clear.]
()
8
if h AP and d N
2
then for some h
AP we have h
AP
h
and
d Rang(d
0
[p
1
(h
)]).
[Why? Like ()
7
.]
Together
there is a sequence h
i
: i < ) such that (for = 1, 2)
(a) it is
AP
-increasing
(b)
if a M
then a B
[h
i
] for some i
(c)
if d N
then d Rang(
d
0
[p
(h
i
)]) for some i < .
From this sequence we can read as isomorphism as required.
4.19
{ps.49}
Conclusion 4.20. If (, , < ) is a T-good,
+
= 2
, M
: <
+
) is an
-increasing continuous sequence of members of EC
(T), M
saturated if is non-
limit and M = M
: <
+
is saturated then for some club E of
+
for
any < < E and , are non-limit, the pairs (M
, M
) and (M
, M
) are
isomorphic.
Proof. By 4.19.
4.20
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

Another form
{ps.56}
Conclusion 4.21. Assume (, , < ) is T-good, e.g. > [T[ is a measurable
cardinality and =
<
. For some function F as in below (as in [Sh:88r, 3.3]),
if M
: <
+
) obeys F see below), then for some club E of
+
(a) cf() M
is saturated
(b) if M
is saturated, cf(
) = and
<
E for = 1, 2 then
(M
1
, M
1
)

= (M
2
, M
2
).
(a) F is a function with domain

M :

M has the form M
i
: i ), a -
increasing continuous sequence such that M
i
is models of T of cardinality
with universe an ordinal [,
+
) and if i is non-limit then M
i
is saturated
(b) F(

M) is such that

MF(

M)) Dom(F)

M = M
: <
+
) obeys F when
+
= sup : F(

M ( + 1) M
+1
.
References
[CK73] Chen C. Chang and H. Jerome Keisler, Model Theory, Studies in Logic and the Foundation
of Math., vol. 73, NorthHolland Publishing Co., Amsterdam, 1973.
[Ehr57] Andrzej Ehrenfeucht, On theories categorical in power, Fundamenta Mathematicae 44
(1957), 241248.
[ER69] Paul Erd os and Richard Rado, Intersection Theorems for Systems of Sets II, Journal of
the London Mathematical Society 44 (1969), 467479.
[KM78] Akihiro Kanamori and Menachem Magidor, The evolution of large cardinal axioms in
set theory, Higher Set Theory, Lecture Notes in Mathematics, vol. 669, Springer Verlag, 1978,
pp. 99275.
[Sh:c] Saharon Shelah, Classication theory and the number of nonisomorphic models, Studies in
Logic and the Foundations of Mathematics, vol. 92, North-Holland Publishing Co., Amsterdam,
xxxiv+705 pp, 1990.
[Sh:3] , Finite diagrams stable in power, Annals of Mathematical Logic 2 (1970), 69118.
[Sh:E53] , Introduction and Annotated Contents, 0903.3598.
[Sh:72] , Models with second-order properties. I. Boolean algebras with no denable auto-
morphisms, Annals of Mathematical Logic 14 (1978), 5772.
[Sh:88r] , Abstract elementary classes near
1
, Chapter I. 0705.4137. 0705.4137.
[Sh:300] , Universal classes, Classication theory (Chicago, IL, 1985), Lecture Notes in
Mathematics, vol. 1292, Springer, Berlin, 1987, Proceedings of the USAIsrael Conference on
Classication Theory, Chicago, December 1985; ed. Baldwin, J.T., pp. 264418.
[Sh:300a] , Stability theory for a model, Chapter V (A), in series Studies in Logic, vol. 20,
College Publications.
[Sh:715] , Classication theory for elementary classes with the dependence property - a
modest beginning, Scientiae Mathematicae Japonicae 59, No. 2; (special issue: e9, 503
544) (2004), 265316, math.LO/0009056.
[Sh:783] , Dependent rst order theories, continued, Israel Journal of Mathematic 173
(2009), 160, math.LO/0406440.
[Sh:863] , Strongly dependent theories, Israel Journal of Mathematics accepted,
math.LO/0504197.
[Sh:868] , When rst order T has limit models, Colloquium Mathematicum submitted,
math.LO/0603651.
[Sh:877] , Dependent T and Existence of limit models, Tbilisi Mathematical Journal sub-
mitted, math.LO/0609636.
[Sh:886] , Denable groups for dependent and 2-dependent theories, Journal of Symbolic
Logic submitted, math.LO/0703045.
[Sh:906] , No limit model in inaccessibles, Proceedings of the Makkaifest accepted,
0705.4131.
(
9
0
0
)

r
e
v
i
s
i
o
n
:
2
0
1
1
-
0
4
-
0
7

m
o
d
i
f
i
e
d
:
2
0
1
1
-
0
4
-
2
6

54 SAHARON SHELAH
[KpSh:946] Itay Kaplan and Saharon Shelah, A counterexample to existence of indiscernibles and
other examples, preprint.
[Sh:950] Saharon Shelah, A dependent dream and recounting types.
Einstein Institute of Mathematics, Edmond J. Safra Campus, Givat Ram, The He-
brew University of Jerusalem, Jerusalem, 91904, Israel, and, Department of Mathe-
matics, Hill Center - Busch Campus, Rutgers, The State University of New Jersey, 110
Frelinghuysen Road, Piscataway, NJ 08854-8019 USA
E-mail address: shelah@math.huji.ac.il
URL: http://shelah.logic.at

Saharon Shelah - Dependent Theories and The Generic Pair Conjecture

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 - Dependent Theories and The Generic Pair Conjecture

Uploaded by

Copyright:

Available Formats

(

) has the same isomorphism type. In

be the -th innite cardinal and the cardinality [U [ of a set U

b a sequence from A and M [= [ a,

(A, M) = tp( a, A, N) : M N, a a sequence of length from N

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

+(T), it is unique up to isomorphism, and the union of an

+(T). On the background so far, see e.g. Chang-Keisler [CK73].

there is a closed unbounded subset E of

: < such that f maps M

(T) equal to one?

(T) = 2 i T is strictly stable (i.e. T is stable, not

(T) = [ + 1[ if T is stable and

(recalling that for a stable T, (T) is cardinal [T[

and its compliment, a M

: (a < x) / p is an end segment.

by the (unary) relation C

in a weak sense, i.e. does not split over B

) realizes the same type over B

d, M, C) in two steps. The rst is to add c of length <

d, M + c, C) is tree like, i.e. if A M C and [A[ < then for some

(T) is the class of models M of T (so M C) of cardinality .

b] belongs to D. Note that if T is dependent, I is a linear order with

(A). Also note that if D is an ultralter on I then Av( a

< (depending on T) such that:

see [Sh:c, Ch.IV,1,3] but F

-constructible over M, see be-

(B) which does not split over A is

) does not split over B

) : < )). We may

with no real change.

-construction A is -full when cf(g(A)) and if q

-constructible over A when there is an F

C and [C[ ([A[ +[T[)

M, (so the complete

< i and k <

is minimal under (a)+(b)+(c).

is well dened i < k. At last we

) and a stationary subset S of S

. Also without loss of generality by 1.2(4) we

. But we could have chosen in stage (1),

clauses (a),(c),(d) of and

) are as required on (N, M).

(A) are dened similarly.

(A) = p S(A) : p is weakly orthogonal to r for every r S

(A) are dened similarly.

(A) i tp( cu, A) S

(A) for every nite u g( c).

: < ). Then we can nd

(N) and = ( x, y) where

(N), see Denition 1.11(3)

b) a formula with parameters from A, see [Sh:783, 1].

< and for u

< and a sequence (

). That is, let p

( y) so for some k < ,

) : < k is a nite subset of p( y)

( y) has cardinality [A[ +[T[ +|M| < .

(T) be saturated and we try to analyze the class of N, M

, which are close enough, in the sense of (, )-limit model.

(M) such that for each p P the model

above has cardinality and is quite dense (e.g. using Fs for = 1, 2)