Sei sulla pagina 1di 87

(

8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES
SH863
Saharon Shelah
The Hebrew University of Jerusalem
Einstein Institute of Mathematics
Edmond J. Safra Campus, Givat Ram
Jerusalem 91904, Israel
Department of Mathematics
Hill Center-Busch Campus
Rutgers, The State University of New Jersey
110 Frelinghuysen Road
Piscataway, NJ 08854-8019 USA
Abstract. We further investigate the class of models of a strongly dependent (rst
order complete) theory T, continuing [Sh 715], [Sh 783] and relatives. Those are prop-
erties (= classes) somewhat parallel to superstability among stable theory, though are
dierent from it even for stable theories. We show equivalence of some of their deni-
tions, investigate relevant ranks and give some examples, e.g. the rst order theory of
the p-adics is strongly dependent. The most notable result is: if |A| +|T| , I C
and |I|
|T|
+() then some J I of cardinality
+
is an indiscernible sequence
over A.
This research was supported by the Israel Science Foundation
I would like to thank Alice Leonhardt for the beautiful typing.
First Typed - 04/May/12
Latest Revision - 09/Sept/4
Typeset by A
M
S-T
E
X
1
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


2 SAHARON SHELAH
Annotated content
0 Introduction, pg.5-7
1 Strong dependent: basic variant, pg.8-28
[We dene
ict
(T) and strongly dependent (= strongly
1
dependent
ict
(T) =

0
), (1.2), note preservation passing from T to T
eq
, preservation under in-
terpretation (1.4), equivalence of some versions of witness <
ict
(T)
(1.5), and we deduce that without loss of generality m = 1 in (1.2). An ob-
servation (1.10) will help to prove the equivalence of some variants. To some
extent, indiscernible sequences can replace an element and this is noted in
1.8, 1.9 dealing with the variant
icu
(T). We end with some examples, in
particular (as promised in [Sh 783]) the rst order theory of the p-adic is
strongly dependent and this holds for similar elds and for some ordered
abelian groups expanded by subgroups. Also there is a (natural) strongly
stable not strongly
2
stable T.]
2 Cutting indiscernible sequence and strongly

, pg.29-42
[We give equivalent conditions to strongly dependent by cutting indiscernibles
(2.1) and recall the parallel result for T dependent. Then we dene
ict,2
(T)
(in 2.3) and show that it always almost is equal to
ict
(T) in 2.8. The ex-
ceptional case is T is strongly dependent but not strongly
2
dependent for
which we give equivalent conditions (2.3 and 2.10.]
3 Ranks, pg.43-53
[We dene M
0

A
M
1
, M
0

A,p
M
2
(in 3.2) and observe some basic prop-
erties in 3.3. Then in 3.5 for most = 1, . . . , 12 we dene <

, <

at
, <

pr
,

,
explicit

-splitting and last but not least the ranks dp-rk
,
(x). Easy prop-
erties are in 3.7, the equivalence of rank is innite, is [T[
+
, T is strongly
dependent in 3.7 and more basic properties in 3.9. We then add more cases
( > 12) to the main denition in order to deal with (verssion of) strongly
dependency and then have parallel claims.]
4 Existence of indiscernibles, pg.54-57
[We prove that if [A[ + [T[ and a


m
C for < Q

+ then for some


u Q

+ of cardinality
+
, a

: u) is indiscernible over A.]


(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 3
5 Concluding Remarks, pg.58-83
[We consider shortly several further relatives of strongly dependent.
(A) Ranks for dependent theories
We redene explicitly

-splitting and dp-rk
,
for more cases, i.e. more s
and for the case of nite

s in a way tting dependent T (in 5.9), point


out the basic equivalence (in 5.9), consider a variant (5.11) and questions
(5.10,5.12).
(B) Minimal theories (or types)
We consider minimality, i.e., some candidates are parallel to
0
-stable theo-
ries which are minimal. It is hoped that some such denition will throw light
on the place of o-minimal theories. We also consider content minimality of
types.
(C) Local ranks for super dependent and indiscernibility
We deal with local ranks, giving a wide family parallel to superstable and
then dene some ranks parallel to those in 3.
(D) Strongly
2
stable elds
We comment on strongly
2
dependent/stable elds. In particular for every
innite non-algebraically closed eld K, Th(K) is not strongly
2
stable.
(E) Strongly
3
dependent
We introduce strong
(3,)
dependent/stable theories and remark on them.
This is related to dimension
(F) Representability and strongly
k
dependent
We dene and comment on representability and

b
t
: t I) being indis-
cernible for I k.
(G) strongly
3
stable and primely regular types.
We prove the density of primely regular types (for strongly
3
stable T) and
we comment how denable groups help.
(H) T is n-dependent
We consider strengthenings n-independent of T is independent.]
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


4 SAHARON SHELAH
0 Introduction
Our motivation is trying to solve the equations x/dependent = superstable/stable
(e.g. among complete rst order theories). In [Sh 783, 3] mainly two approximate
solutions are suggested: strongly

dependent for = 1, 2; here we try to investigate


them not relying on [Sh 783, 3]. We dene
ict
(T) generalizing (T), the denition
has the form <
ict
(T) i there is a sequence
i
( x, y
i
) : i < ) of formulas such
that ....
Now T is strongly dependent (= strongly
1
dependent) i
ict
(T) =
0
; prototyp-
ical examples are: the theory of dense linear orders, the theory of real closed elds,
the model completion of the theory of trees (or trees with levels), and the theory
of the p-adic elds (and related elds with valuations). (The last one is strongly
1
not strongly
2
dependent, see 1.17.)
For T superstable, if a
t
: t I) is an indiscernible set over A and C is nite
then for some nite I

I, a
t
: t II

) is indiscernible over AC; moreover over


AC a
t
: t I

. In 2 we investigate the parallel here, when I is a linear order,


complete for simplicity (as in [Sh 715, 3] for dependent theories). But we get two
versions: strongly

dependent = 1, 2 according to whether we like to generalize


the rst version of the statement above or the moreover.
Next, in 3, we dene and investigate rank, not of types but of related objects
x = (p, M, A) where, e.g. p S
m
(M A); but there are several variants. For some
of them we prove T is strongly dependent i the rank is always < i the rank
is bounded by some < [T[
+
. We rst deal with the ranks related to strongly
1
dependent and then for the ones related to strongly
2
dependent.
Further evidence for those ranks being of interest is in 4 where we use them to
get indiscernibles. Recall that if T is stable, [A[ =
|T|
, a

C for < :=
+
then for some stationary S , a

: S) is indiscernible over A, [S[ = ,


we can write this as ()
<
T,
. We can get similar theorems from set theoretic
assumptions: e.g. a measurable cardinal, very interesting and important but not
for the present model theoretic investigation.
We may wonder: Can we classify rst order theories by
T
()

, as was asked
by Grossberg and the author (see on this question [Sh 702, 2.9-2.20]). A positive
answer appears in [Sh 197], but under a very strong assumption on T: not only T
is dependent but every subsets P
1
, . . . , P
n
of [M[ the theory Th(M, P
1
, . . . , P
n
) is
dependent, i.e., being dependent is preserved by monadic expansions.
Here we prove that if T is strongly stable and [T[ then Q

+
T
(
+
)
<

+
.
We certainly hope for a better result (using Q
n
([T[) for some x n or even (2

)
+
instead of Q

+) and weaker assumptions, say T is dependent (or less) instead


T is strongly dependent. But still it seems worthwhile to prove 4.1 particularly
having waited so long for something.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 5
Let strongly

stable mean strongly

dependent and stable. As it happens (for


T), being superstable implies strong
2
stable implies strong
1
stable but the inverses
fail. So strongly

dependent does not really solve the equation we have started with.
However, this is not necessarily bad, the notion strongly

stable seems interesting


in its own right; this applies to the further variants.
We give a simplest example of a theory T which is strongly
1
stable and not
strongly
2
stable in the end of 1 as well as prove that the (theories of the) p-adic
eld is strongly stable (for any prime p) as well as similar enough elds.
In 5 we comment on some further properties and ranks. Such further properties
hopefully will be crucial in [Sh:F705], if it materializes; it tries to deal mainly with
K
or
-representable theories and contain other beginning as well. We comment on
ranks parallel to those in 3 suitable for all dependent theories.
We further try to look at theories of elds. Also we deal with the search for
families of dependent theories T which are unstable but minimal, much more
well behaved. For many years it seems quite bothering that we do not know how
to dene o-minimality as naturally arising from a parallel to stability theory rather
than as an analog to minimal theories or generalizes examples related to the theory
of the eld of the reals and its expansions. Of course, the answer may be a somewhat
larger class. This motivates Firstenberg-Shelah [FiSh:E50] (on Th(1), specically
on perpendicularly is simple), and some denitions in 5. Another approach to
this question is of Onshuus in his very illuminating works on th-forking [On0x1]
and [On0x2].
A result from [Sh 783, 3,4] used in [FiSh:E50] says that
0.1 Claim. Assume T is strongly
2
dependent
(a) if G is a denable group in C
T
and h is a denable endomorphism of G
with nite kernels then h is almost onto G, i.e., the index (G : Rang(h)) is
nite
(b) it is not the case that: there are denable (with parameters) subset (C, a
1
)
of C, an equivalence relation E
a
2
= E(x, y, a
2
) on (C, a
1
) with inn-
itely many equivalence classes and (x, y, z, a
3
) such that E(c, c, a
2
)
(x, y, c, a
3
) is a one-to-one function from (a co-nite subset of ) (C, a
1
)
into c/E
a
2
.
We continue investigating dependent theories in [Sh:900], [Sh 877], [Sh:906], more
recently [Sh:F931] and Kaplan-Shelah and concerning denable groups in [Sh 876],
[Sh 886].
We thank Moran Cohen, Itay Kaplan, Aviv Tatarski and a referee for pointing
out deciencies.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


6 SAHARON SHELAH
0.2 Notation. 1) Let
t
be if t = 1 or t = true and if t = 0 or t = false.
2) S

(A, M) is the set of complete types over A in M (i.e. nitely satisable in M)


in the free variables x
i
: i < ).
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 7
1 Strongly dependent: basic variant
1.1 Convention. 1) T is complete rst order xed.
2) C = C
T
a monster model for T.
Recall from [Sh 783]:
1.2 Denition. 1)
ict
(T) =
ict,1
(T) is the minimal such that for no =

i
( x, y
i
) : i < ) is

consistent with T for some ( every) where


g( x) = m, g( y
i
m
) = g( y
i
) and

=
i
( x

, y
i

)
if((i)=)
:

, < and i < .
1A) We say that =
i
( x, y
i
) : i < ) witness <
ict
(T) (with m = g( x)) when
it is as in part (1).
2) T is strongly dependent (or strongly
1
dependent) when
ict
(T) =
0
.
Easy (or see [Sh 783]):
1.3 Observation. If T is strongly dependent then T is dependent.
1.4 Observation. 1)
ict
(T
eq
) =
ict
(T).
2) If T

= Th(M

) for = 1, 2 then
ict
(T
1
)
ict
(T
2
) when:
() M
1
is (rst order) interpretable in M
2
.
3) If T

= Th(C, c)
cA
then
ict
(T

) =
ict
(T).
4) If M is the disjoint sum of M
1
, M
2
(or the product) and Th(M
1
), Th(M
2
) are
dependent then so is Th(M); so M
1
, M
2
, M has the same vocabulary.
5) In Denition 1.2, for some ,

is consistent with T i for every ,


is consis-
tent with T.
Remark. 1) Concerning Part (4) for strongly dependent, see Cohen-Shelah [CoSh:E65].
Proof. Easy.
1.4
1.5 Observation. Let g( x) = m; =
i
( x, y
i
) : i < ) and let

i
( x, y

i
) : i <
) where

i
( x, y

i
) = [
i
( x, y
1
i
)
i
( x, y
2
i
)] and let

i
( x, y

i
) : i < ) where
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


8 SAHARON SHELAH

i
( x, y

i
) = [
i
( x, y
1
i
)
i
( x, y
2
i
)]. Then
1


2


3

(

2)
2

[]

(

2)
3

[]
and


for = 2, 3 and
3


1


1

where

[]
=
i
( x, y
i
)
(i)
: i < ) and

1

witness <
ict
(T)

2

we can nd a
i
k
: k < , i < ) in C such that g( a
i
k
) = g( y
i
), a
i
k
: k <
) is indiscernible over a
j
k
: j < , j ,= i, k < for each i < and

i
( x, a
i
0
)
i
( x, a
i
1
) : i < is consistent, i.e. nitely satisable in C

3

like
2

but in the end

i
( x, a
i
0
)
i
( x, a
i
1
) : i < is consistent.
1.6 Remark. 1) We could have added the indiscernibility condition to
1

, i.e., to
1.2(1) as this variant is equivalent to
1

.
2) Similarly we could have omitted the indiscernibility condition in
2

but demand
in the end: if k

<
i
< for i < then
i
( x, a
i
k
i
)
i
( x, a
i

i
) : i < is
consistent and get an equivalent condition.
3) Similarly we could have omitted the indiscernibility condition in
3

but demand
in the end if k
i
<
i
< for i < then
i
( x, a
i
k
i
)
i
( x, a
1

i
) : i < is
consistent and get an equivalent condition.
4) We could add
3


1

.
5) In
2

,
3

(and the variants above) we can replace by any ), see 1.7).
6) What about
2


1

? We shall now describe a model whose theory is a
counterexample to this implication. We dene a model M,
M
= P, P
i
, R
i
: i <
, P a unary predicate, P
i
a unary predicate, R
i
a binary predicate:
(a) [M[ the universe of M is ( )

(b) P
M
=

(c) P
M
i
= i
(d) R
M
i
= (, (i, q)) :

, q and [= (i) q
(e)
i
(x, y) = P(x) P
i
(y) R
i
(x, y) for i < .
Now
() Why (for Th(

M)) do we have
2

?
For i < , k < let a
i
k
= (i, k) P
M
i
recalling .
Easily a
i
k
: k < , i < ) are as required in
2

. E.g. the unique

realizing
the type. Also for each i < , the sequence a
i
k
: k < ) is indiscernible over
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 9
a
j
m
: j < , j ,= i and m < .
Why? Because for every automorphism of the rational order (, <), for the given
i < we can dene a function
i
with domain M by
()
1
for j < and q we let
i
((j, q)) be (j, q) if j ,= i and (j, (q)) if j = i
()
2
for ,

we have
i
() = i (j < )(j ,= i (j) = (j)) and
(i) =
i
((i)).
So
i
is an automorphism of M over
_
j=i
P
M
j
which includes the function (a
i
q
, a
i
(q)
) :
q
() Why (for Th(M)), we do not have
1

?
Because M [= (y
1
, y
2
)[P
i
(y
1
) P
i
(y
2
) y
1
,= y
2

2

=1
(x)(
i
(x, y

)
P(x)
i
(x, y
3
))].
Proof. The following series of implications clearly suces.

1

implies
2

Why? As
1

, clearly for any
0
we can nd a
i


g( y
i
)
C for i < , < and
c

:

), c


g( x)
C such that [=
i
[ c

, a
i

] i (i) = . By some applications


of Ramsey theorem (or polarized partition relations) without loss of generality a
i

:
< ) is indiscernible over a
j

: j < , j ,= i, < for each i < . Now those


a
i

s witness
2

as c

witness the consistency of the required type when



0.

2


3

(hence in particular
2


3

and
2


3

).
Trivial; read the denitions.

3


2

(hence in particular
3


3

and
2


3

).
By compactness, for the dense linear order 1 we can nd a
i
t
for i < , t 1
such that for each i < the sequence a
i
t
: t 1) indiscernible over a
j
s
: j ,=
i, j < , s 1 and for any s
0
<
R
s
1
the set
i
( x, a
i
s
0
)
i
( x, a
i
s
1
) : i < is
consistent, say realized by c = c
s
0
,s
1
. Now let u = i < : C [=
i
[ c, a
i
s
0
] and
for n < dene

b
i
n
as a
i
s
0
+n(s
1
s
0
)
if i u and as a
i
s
1
n(s
1
s
0
)
if i u. Now

b
i
n
: n < , i < ) exemplies
2

.

2

implies
1

(hence by the above
2


2

and
3


3

).
Let a
i

: < , i < ) witness


2

and c realizes
i
( x, a
i
0
)
i
( x, a
i
1
) : i < .
Without loss of generality a
i
t
is well dened for every t Z not just t , (and
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


10 SAHARON SHELAH
i < ), and a
i
t
: t Z) is an indiscernible sequence over a
j
s
: j i and
s Z. Also without loss of generality for each i < , a
i

: [2, )) as well as
a
i
1n
: n ) are indiscernible sequences over a
j
t
: j < , j ,= i and t Z c.
For t Z, i < let

b
i
t
= a
i
2t
a
i
2t+1
, so C [=

i
[ c,

b
i
0
] (as this just means C [=

i
( c, a
i
0
)
i
[ c, a
i
1
]) and C [=

i
[ c,

b
i
s
] when s Z0 (as the sequences c a
i
2s
and c a
i
2s+1
realize the same type). So

b
i

: < , i < ) witness


1

.

3

implies
3

.
Read the denitions.

3

implies that for some

2 we have
1

[]
.
As in the proof of
2


1

; but we elaborate: let

a
i

b
i

: < ) : i <
_
witness
3

noting

i
( x, y
i
1
, y
i
2
) : i < ) where g( y
i
1
) = g( y
i
) = g( y
i
2
). Let
c realize

i
( x, a
i
0
,

b
i
0
)

i
( x, a
i
1
,

b
i
1
) : i < . Without loss of generality for each
i < the sequence a
i

b
i

: 2 < ) is indiscernible over a


j

b
j

: j i
and < c.
By this extra indiscernibility assumption for each i < we can nd
0
(i),
1
(i)
0, 1 such that n 2 C [=
i
[ c, a
i
n
]

0
(i)

i
[ c,

b
i
n
]

1
(i)
. By the choice of c we
have C [=

i
( c, a
i
0
,

b
i
0
)

i
( c, a
i
1
,

b
i
1
), hence by the choice of

i
, we cannot have
C [=
i
[ c, a
i
0
]

0
(i)

i
[ c, a,

b
i
0
]

1
(i)

i
[ c, a
i
1
]

0
(i)

i
[ c,

b
i
1
]

1
(i)
.
Hence there are
3
(i),
4
(i) 0, 1 such that

4
(i) = 0 C [=
i
[ c, a
i

3
(i)
]
1
0
(i)

4
(i) = 1 C [=
i
[ c,

b
i

3
(i)
]
1
1
(i)
.
Lastly choose = 1

4
(i)
(i) : i < ) and we choose

d
i

: < , i < ) as follows:


if
4
(i) = 0 and n = 0 then

d
i
n
= a
i

3
(i)
if
4
(i) = 0 and n > 0 then

d
i
n
= a
i
1+n
if
4
(i) = 1 and n = 0 then

d
i
n
=

b
i

3
(i)
if
4
(i) = 1 and n > 0 then

d
i
n
=

b
i
1+n
.
Now check that

d
i

: < and i < ) witness


1

[]
.

3

[]
,
3

are equivalent where

2.
Why? Because the formula (
i
(x, a
i
0
)
i
(x, a
i
1
)) is equivalent to (
i
(x, a
i
0
)
(i)

i
(x, a
i
1
)
(i)
.
1.5
1.7 Observation. 1) In Denition 1.2 without loss of generality m(= g( x)) is 1.
2) For any we have: <
ict
(T) i for some innite linear order I
i
(for i < )
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 11
and a
i
t
: t I
i
, i < ) such that a
i
t
: t I
i
) is indiscernible over a
j
s
: s I
j
and
j ,= i, j < A and nite C, for ordinals i < , the sequence a
i
t
: t I
i
) is not
indiscernible over A C.
3) In 1.5, for any (
0
) from the statement
2

we get an equivalent one if we
replace by ; similarly for
3

.
Proof. 1) For some m, there is =
i
( x, y
i
) : i < ), g( x) = m witnessing
<
ict
(T); without loss of generality m is minimal. Fixing by 1.5 we know that

2

from that observation 1.5 hold. Let a
i

: i < , < ) exemplify


2

with
instead and let c = c
i
: i < m) realize
i
( x, a
i
0
)
i
( x, a
i
1
) : i < .
Case 1: For some u , [u[ < for every i u the sequence a
i

: < ) is an
indiscernible sequence over a
j

: j ui c
m1
.
In this case for i u let
i
( x

, y

i
) :=
i
( x (m 1), x
m1
) y
i
) and

=

i
( x

, y

i
) : i u) and

b
i

= c
m1
) a
i

for < , i u and =


i
( x

, y

i
) :
i u). Now

b
i

: < , i u) witness that (abusing our notation)


2

holds
(the consistency exemplied by c (m 1)), hence (in the notation of 1.5)
1

[]
holds for some
\u
2 contradiction to the minimality of m.
Case 2: Not Case 1.
We choose v

by induction on < such that

(a) v

: <
(b) v

is nite
(c) for some i v

, a
i

: < ) is not indiscernible over


a
j

: j v

i, < c
m1

(d) under (a) + (b) + (c), [v

[ is minimal.
In the induction step, the set u

= v

: < cannot exemplify case 1, so


for some ordinal i() u

the sequence a
i()

: < ) is not indiscernible


over a
j

: j u

i() and < c


m1
, so by the nite character of
indiscernibility, there is a nite v u

i() such that a


i()

: < ) is not
indiscernible over a
j

: j v, < c
m1
. So v

= i() v satises
(a) + (b) + (c) hence some nite v

satises clauses (a), (b), (c) and (d).


Having carried the induction let i

() v

exemplify clause (c). We can nd a


sequence

d

from a
j

: j v

() and < such that a


i

()

: < ) is
not indiscernible over c
m1
)

.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


12 SAHARON SHELAH
Also we can nd n() < and ordinals
,,0
<
,,1
< . . . <
,,n()1
< for
= 1, 2 such that the sequences

d a
i

()

,1,0
. . . a
i

()

,1,n()1
and

d a
i

()

,2,0
. . . a
i

()

,2,n()1
realize dierent types over c
m1
.
Now we consider a
i

()

. . . a
i

()
+n()1
where := max
,1,n()1
+1,
,2,n()1
+
1, so renaming without loss of generality
,1,n()1
<
,2,0
. Omitting some
a
i

()

s without loss of generality

,1,m
= m,
,2,m
= n() + m for m < n().
Now we dene

b

:=

d

a
i

()
n()
. . . a
i

()
n()+n()1
for < , < .
By the indiscernibility of a
i

()

: < ) over

d

a
j

: j v

, <
a
j

: j i

(), < we can deduce that

: < ) is an indiscernible
sequence over

: and < . But by an earlier sentence

0
,

1
realizes
dierent types over c
m1
so we can choose

(x, y

) such that C [=

(c
m1
,

0
)

i
(c
m1
,

b
i
1
) for i < .
So

: < , < ) and

(x, y

) : < ) satisfy the demands on


a
i
k
: k < , i < ),
i
(x, y
i
) : i < ) in
2

for m = 1 (by 1.5s notation), so by 1.5
also
1

[]
holds for some

2 so we are done.
2) Implicit in the proof of part (1) (and see case 1 in the proof of 2.1).
3) Trivial.

1.7
A relative of
ict
(T) is
1.8 Denition. 1)
icu
(T) =
icu,1
(T) is the minimal such that for no m <
and =
i
( x
i
, y
i
) : i < ) with g( x
i
) = m n
i
can we nd a
i


g( y
i
)
C for
< , i < and c
,n

m
C for

such that:
(a) c
,n
: n < ) is an indiscernible sequence over a
i

: < , i <
(b) for each

and i < we have C [=
i
( c
,0
. . . c
,n
i
1
, a
i

)
if(=(i))
.
2) If is as in (1) then we say that it witnesses <
icu
(T).
3) T is strongly
1,
dependent if
icu
(T) =
0
.
1.9 Claim. 1)
icu
(T)
ict
(T).
2) If cf() >
0
then
icu
(T) >
ict
(T) > .
3) The parallels of 1.4, 1.5, 1.7(2) hold
1
.
1
and of course more than 1.7(2), using an indiscernible sequence of m

-tuples, for any m

< .
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 13
Proof. 1) Trivial.
2) As in the proof of 1.7.
3) Similar.
1.9

To translate statement on several indiscernible sequences to one (e.g. in 2.1) notes:
1.10 Observation. Assume that for each < , I

is an innite linear order, the


sequence a
t
: t I

) is indiscernible over A a
t
: t I

and (and
for notational simplicity I

: < ) are pairwise disjoint) and let I = I

: <
, t I

g( a
t
) = () and lastly for we let () = () : < .
Then there is

b
t
: t I) such that
(a) g(

b
t
) = ()
(b)

b
t
: t I) is an indiscernible sequence over A
(c) t I

a
t
=

b
t
[

)
(d) if C C and P is a set of cuts of I such that [J is a convex subset of
I not divided by any member of P

b
t
: t J) is indiscernible over
A C] then we can nd P

: < ), P

is a set of cuts of I

such that
[P

[ : < = [P[ and if < , J is a convex subset of I

not divided
by any member of P

then a
t
: t J) is indiscernible over A C
(e) if C C and P is a set of cuts of I such that [J is a convex subset of
I not divided by any member of P

b
t
: t J) is indiscernible over
A C b
s
: s IJ] then we can nd P

: < ), P

is a set of cuts
of I

such that [P

[ : < = [P[ and if < , J is a convex subset


of I

not divided by any member of P

then a
t
: t J) is indiscernible
over A C a
t
: t IJ
(f) moreover in clause (d),(e) we can choose P

as the set of non-trivial cuts


of I

induced by P, i.e.(J

, J

) : (J

, J

) P(I

, ), (, I

).
Proof. Straightforward. E.g.,
Without loss of generality I

: < ) are pairwise disjoint and let I = I

:
< . We can nd

b

t

()
C for t I, < such that: if n < ,
0
< . . . <

n1
< , t

0
<
I
. . . <
I
t

1
and s

0
<
I

. . . <
I
a

1
for < n then the sequence
(

0
t
0
0
. . .

0
t
0
k
0
1
) . . . (

n1
t
n1
0
. . .

n1
t
n1
k
n1
1
) realizes the same type as the sequence
( a

s
0
0
. . . a

0
s
0
k
n
1
) . . . ( a

n1
s
n1
0
. . . a

n1
s
n1
k
n1
1
); this is possible by compactness. Us-
ing an automorphism of C without loss of generality t I

t
= a

t
. Now for
t I let a

t
be ( a
0
t
a
1
t
. . . a
1

. . . )
<
.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


14 SAHARON SHELAH
Clauses (a)+(b)+(c) trivially hold and clauses (d),(e),(f) follows.
1.10

In the following we consider natural examples which are strongly dependent, see
more in 2.5.
1.11 Claim. 1) Assume T is a complete rst order theory of an ordered abelian
group expanded by some individual constants and some unary predicates P
i
(i < i())
which are subgroups and T has elimination of quantiers.
T is strongly dependent i we cannot nd i
n
< i() and
n
Z0 for n < such
that:
() we can nd b
n,
C for n, < such that
(a)
1
<
2

n
(b
n,
2
b
n,
1
) / P
C
i
n
(b) for every

there is c

such that c

b
n,(n)
P
C
i
n
for n < .
2) Let M be (Z, +, , 0, 1, <, P
n
) where P
n
= na : a Z so we know that
T = Th(M) has elimination of quantiers. Then T is strongly dependent hence
Th(Z, +, , 0, <) is strongly dependent.
1.12 Remark. 1) This generalizes the parallel theorem for stable abelian groups.
2) Note, if G is the ordered abelian group with sets of elements Z[x], addition of
Z[x] and p(x) > 0 i the leading coecient is > 0, in Z, P
n
as above (so denable),
then Th(G) is not strongly dependent using P
n
for n prime.
2) On elimination of quantiers for ordered abelian groups, see Gurevich [Gu77].
Proof. 1) The main point is the if direction. We use the criterion from 2.1(2),(4)
below. So let a
t
: t I) be an innite indiscernible sequence and c C (with a
t
not
necessarily nite). Without loss of generality C [= c > 0 and a
t
= a
t,
: <

)
list the members of M
t
, a model and even a [T[
+
-saturated model, (see 2.1(4)) and
let p
t
= tp(c, M
t
).
Note that
()
1
if a
s,i
= a
t,j
and s ,= t then a
r,i
: r I) is constant.
Obviously without loss of generality c / M
t
: t I but C is torsion free (as an
abelian group because it is ordered) hence
()
2
Z0 c / M
t
: t I
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 15
()
3
for t I, a M
t
and Z0 let

a

i()+1
2 be such that [

a
(i()) =
1 c > a] and for i < i(), [

a
(i) = 1 c a P
C
i
]
()
4
for t I and a M
t
let p
a
:=
_
Z\{0}
(p

a
q

a
) where
2
p

a
(x) := x ,=
a, (x > a)

a
(i())
and q

a
(x) := P
i
(x a)

a
(i)
: i < i().
Now

0
for Z0 and <

let I

= t I : a
t,
< c

1
u
1
, u
0
, u
1
) is a partition of

where
(a) u
1
= <

: for every s <


I
t we have C [= a
t,
< a
s,

(b) u
0
= <

: for every s <


I
t we have C [= a
s,
= a
t,

(c) u
1
= <

: for every s <


I
t we have C [= a
s,
< a
t,

2
if Z0 then
(a) I

is an initial segment of I when u


1
(b) I

is an end segment of I when u


1
(c) I

, I when u
0
(d) I

: u
1
, I has at most 2 members.
[Why? Recall <
C
is a linear order. So for each Z0, u
1
by the denition
of u
1
the set I

:= t I : a
t,
< c is an initial segment of I, also t II


c <
C
a
t,
as c / M
s
: s I by ()
2
.
Now suppose , u
1
and [I

[ > 1 and I

, I

/ , I then choose t
1
<
I
t
2
from I

and t
0
I

, t
3
II

. As I

, I

are initial segments and t


0
<
I
t
1
<
I
t
2
<
I
t
3
, necessarily C [= a
t
0
,
< c < a
t
1
,
a
t
2
,
< c < a
t
3
,
. If a
t
1
,

C
a
t
2
,
we can deduce a contradiction (C [= c < a
t
1
,
a
t
2
,
< c). Otherwise by the
indiscernibility of the sequence (a
t,
, a
t,
) : t I) we get C [= a
t
3
,
< a
t
0
,
and a
similar contradiction. So [I

[ 1.
So I

, I

/ , I [I

[ 1 and by symmetry [I

[ 1. So [I

:
u
1
, I[ 2, i.e. clause (d) of
2
holds; the other clauses should be clear.]
Now clearly

3
if , < (), Z0 and a
t,
= a
t,
(for some equivalently for every
t I) then:
(a) ( u
1
) ( u
1
)
2
recall that
1
= ,
0
=
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


16 SAHARON SHELAH
(b) ((c) < a
t,
) (a
t,
< (()c)) recalling c, ()c /
_
tI
M
t
(c) I

= II

.
Also

4
if
1
,
2
are from 1, 2, . . . and
1
a
t,
=
2
a
t,
then
(a) [ u
1
u
1
], [ u
0
u
0
] and [ u
1
u
1
]
(b) (t I

) (t I

) hence I

= I

.
[Why? Clause (a) is obvious. For clause (b) note that t I

a
t,
<
2
c

1
a
t,
<
1
(
2
c)
2
a
t,
<
2
(
1
c) a
t,
<
1
c t I

.]
By symmetry, i.e. by
3
clearly

5
the statement (c),(d) in
2
holds for u
1
.
Obviously

6
if u
0
then I

, I.
Together

7
I

: <

and Z0, I hence has 4 members.


Hence

0
there are initial segments J

of I for < () 4 such that: if s, t belongs


to I and < () [s J

t J

] then

a
t,
(i()) =

a
s,
(i()).
[Why? By the above and the denition of

a
t,
(i()) we are done.]

1
for each t I we have p
a
(x) : a M
t
p
t
(x).
[Why? Use the elimination of quantiers and the closure properties of
M
t
. That is, every formula in p
t
(x) is equivalent to a Boolean combi-
nation of quantier free formulas. So it suces to deal with the cases
(x, a) p
t
(x) which is atomic or negation of atomic and x appear. As
for b
1
, b
2
C exactly one of the possibilities b
1
< b
2
, b
1
= b
2
, b
2
< b
1
holds
and by symmetry, it suces to deal with
1
(x, a) >
2
(x, a),
1
(x, a) =

2
(x, a), P
i
((x, a)), P
i
((x, a)) where (x, y),
1
(x, y),
2
(x, y) are terms
in '(
T
). As we can substract, it suces to deal with (x, a) > 0, (x, a) =
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 17
0, P
i
((x, a)), P
i
((x, a)). By linear algebra as M
t
is closed under the op-
erations, without loss of generality (x, a) = x a
t,
for some Z and
<

, and without loss of generality ,= 0. The case (x) = (x a


t,
=
0) p(x) implies c M
t
(as M is torsion free) which we assume does
not hold. In the case (x, a) = (x a
t,
> 0) use p

a
t,
(x), in the case
(x, a) = P
i
(x a
t,
) or (x, a) = P
i
(x a
t,
) use q

a
t,
(x) for

a
t,
(i).]

2
if Z0, n < and a
0
, . . . , a
n1
M
t
then for some a M
t
we have
< n i < i()

(i) = 1

a
(i) = 1
[Why? Let a

M
t
realize p
t
a
0
, . . . , a
n1
, exist as M
t
was chosen as
[T[
+
-saturated; less is necessary. Now c a

P
C
i
a

P
C
i

(c a

) = ((c a

) (a

)) P
C
i
and let a := a

.]

3
assume Z0, i < i(), <

, s
1
<
I
s
2
and t Is
1
, s
2
then:
(a) if

a
s
1
,
(i) = 1 and

a
s
2
,
(i) = 0 then

a
t,
(i) = 0
(b) if

a
s
1
,
(i) = 0 and

a
s
2
,
(i) = 1 then

a
t,
(i) = 0.
[Why? As we can invert the order of I it is enough to prove clause (a).
By the choice of a

a
we have c a
s
1
,
P
C
i
, c a
s
2
,
/ P
C
i
hence
a
s
1
,
a
s
2
,
/ P
C
i
hence also a
s
2
,
a
s
1
,
/ P
C
i
.
By the indiscernibility we have a
t,
a
s
1
,
/ P
C
i
and as c a
s
1
,
P
C
i
we
can deduce c a
t,
/ P
C
i
hence

a
t,
(i) = 0. So we are done.]

4
for each Z0, i < i() and <

the set I

i,
:= t :

a
t,
(i) = 1 is
, I or a singleton.
[Why? By
3
.]

5
if I

= I

i,
: Z0, i < i(), <

and I

i,
is a singleton is innite
then (possibly inverting I) we can nd t
n
I and
n
<

,
n
Z0 and
i
n
< i() for n < such that
(a) t I then [
n
c a
t,
n
P
C
i
n
] t = t
n
for every n <
(b) a
t,
n
a
s,
n
: s ,= t I) are pairwise not equal mod P
C
i
n
(c) t
n
< t
n+1
for n < .
[Why? Should be clear.]

6
if I

= I

i,
: Z0, <

, i < i() and I

i,
is a singleton is nite
and J

( < () 6) are as in
0
, then tp( a
s
, c) = tp( a
t
, c) whenever
(s, t II

<()
(s J

t J

) recalling a
t
list the elements of M
t
.
[Why? By
4
and
1
(and
0
) recalling the choice of p
a
in ()
4
.]
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


18 SAHARON SHELAH
Assume c, a
t
: t I) exemplify T is not strongly dependent then I

cannot be
nite (by
6
) hence I

is innite so by
5
we can nd (t
n
,
n
,
n
, i
n
) : n < ) as
there.
That is, for n < , < let b
n,
:= a
t

,
n
. So

7

n
c b
n,
P
C
i
n
i
n
c a
t

,
n
P
C
i
n
i t

= t
n
i = n

8
if
1
<
2
then b
n,
2
b
n,
2
/ P
C
i
n
.
[Why? By clause (b) of
5
.]
Now

9
if

is increasing then there c

C such that
n <
n
c

b
n,(n)
P
C
i
n
.
[Why? As a
t
: t I) is an indiscernible sequence, there is an automorphism
f = f

of C which maps a
t
n
to a
t
(n)
for t I so f

(b
,n
) = b
n,(n)
. Hence
c

= f

(c) satises n <


n
f(c) b
,(n)
P
C
i
n
.]
Now b
n,
: n, < ) almost satises () of 1.11. Clause (a) holds by
8
and
clause (b) holds for all increasing

. By compactness we can nd

n,
: n, <
) satisfying (a) + (b) of () of 1.11.
[Why? Let = P
i
n
(
n
x

y
n,(n)
) :

, n < P
i
n
(
n
x
n,
1

n
x
n,
2
) :
n < ,
1
<
2
< . If is satised in C we are done, otherwise there is a nite
inconsistent

, let n

be such that: if y
n,
appear in

then n, < n

. But
the assignment y
n,
b
nn

+
for n < n

, < n

exemplied that

is realized, so
we have proved half of the claim. The other direction should be clear, too.]
2) The rst assertion (on T) holds by part (1); the second holds as the set of terms
0, 1, 2, . . . , n1 is provably a set of representatives for Z/P
n
which is nite.
1.11
1.13 Example: Th(M) is not strongly stable when M satises:
(a) has universe

(b) is an abelian group as a power of (, +),


(c) P
M
n
= f M : f(n) = 0, a subgroup.
We now consider the p-adic elds and more generally valued elds.
1.14 Denition. 1) We dene a valued eld M as one in the Denef-Pas language,
i.e., a model M such that:
(a) the elements of M are of three sorts:
() the eld P
M
0
which (as usual) we call K
M
, so K = K
M
is the eld of M
and has universe P
M
0
so we have appropriate individual constants (for
0, 1), and the eld operations (including the inverse which is partial)
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 19
() the residue eld P
M
1
which (as usual) is called k
M
, so k = k
M
is a eld
with universe P
M
1
so with the appropriate 0, 1 and eld operations
() the valuation ordered abelian group P
M
2
which (as usual) we call
M
,
so =
M
is an ordered abelian group with universe P
M
2
so with 0,
addition, subtraction and the order
(b) the functions (and individual constants) of K
M
, k
M
,
M
and the order of

N
(actually mentioned in clause (a))
(c) val
M
: K
M

M
, the valuation
(d) ac
M
: K
M
k
M
, the function giving the leading coecient (when as in
natural cases the members of K are power series)
(e) of course, the sentences saying that the following hold:
()
M
is an ordered abelian group
() k is a eld
() K is a eld
() val,ac satises the natural demands.
1A) Above we replace language by -language when: in clause (b), i.e. (a)(),

M
has 1

(the minimal positive elements) and we replace (d) by


(d)

ac
M
n
: K
M
k
M
satises:

<n
ac
M

(x) = ac
M
k
(y) val
M
(x y) >
val
m
(x) +n.
2) We say that such M (or Th(M)) has elimination of the eld quantier when:
every rst order formula (in the language of Th(M)) is equivalent to a Boolean
combination of atomic formulas, formulas about k
M
(i.e., all variable, free and
bounded vary on P
M
1
) and formulas about
M
; note this denition requires clause
(d) in part (1).
It is well known that (on 1.15,1.16 see, e.g. [Pa90], [CLR06]).
1.15 Claim. 1) Assume is a divisible ordered abelian group and k is a perfect eld
of characteristic zero. Let K be the xed power series for (, k), i.e. f : f

k
and supp(f) is well ordered where supp(f) = s : f(s) ,= 0
k
. Then the model
dened by (K, , k) has elimination of the eld quantiers.
2) For p prime, we can consider the p-adic eld as a valued eld in the Denef-Pas
-language and its rst order theory has elmination of the eld quantiers (this
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


20 SAHARON SHELAH
version of the p-adics and the original one are (rst-order) bi-interpretable; note
that the eld k here is nite and formulas speaking on which is the ordered abelian
group Z are well understood).
We will actually be interested only in valuation elds M with elimination of the
eld quantiers. It is well known that
1.16 Claim. Assume C = C
T
is a (monster,i.e. quite saturated) valued eld in the
Denef-Pas language (or in the -language) with elimination of the eld quantiers.
If M C then
(a) it satises the cellular decomposition of Denef which implies
3
:
if p S
1
(M) and P
0
(x) p then p is equivalent to
p
[]
:= p
[]
c
: c P
M
0
where p
[]
c
= p
[,1]
c
p
[,2]
c
and
p
[,1]
c
= (val(x c),

d) p : (x, y) is a formula speaking on
M
only so

d
M
, c P
M
0
and
p
[,2]
c
= (ac(x c),

d) p : speaks on k
M
only but for the -language
we should allow (ac
0
(x c), . . . , ac
n
(x c),

d) for some n <
(b) if p S
1
(M), P
0
(x) p and c
1
, c
2
P
M
0
and val
M
(xc
1
) <

M
val
M
(xc
2
)
belongs to p(x) then p
[]
c
2
(x) p
[]
c
1
(x) and even val(x c
1
) < val(x c
2
)
p
[]
c
1
(x)
(c) for c
>
(k
M
), the type tp( c, , k
M
) determines tp( c, , M) and similarly
for
M
.
1.17 Claim. 1) The rst order theory T of the p-adic eld is strongly dependent.
2) For any theory T of a valued eld F which has elimination of the eld quantier
we have:
T is strongly dependent i the theory of the valued ordered group and the theory of
the residue elds of F are strongly dependent.
3) Like (2) when we use the -language and we assume k
M
is nite.
1.18 Remark. 1) In 1.17 we really get that T is strongly dependent over the residue
eld + the valuation ordered abelian group.
2) We had asked in a preliminary version of [Sh:783,3]: show that the theory of
the p-adic eld is strongly dependent. Udi Hrushovski has noted that the criterion
(St)
2
presented there (and repeated in 0.1 here from [Sh 783, 3.10=ss.6]) apply so
3
note: p S
1
(A, M), A M is a little more complicated
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 21
T is not strongly
2
dependent. Namely take the following equivalence relation E on
Z
p
:val(xy) val(c), where c is some xed element with innite valuation. Given
x, the map y (x +cy) is a bijection between Z
p
and the class x/E.
3) By [Sh 783, 3], the theory of real closed elds, i.e. Th(1) is strongly dependent.
Onshuus shows that also the theory of the eld of the reals is not strongly
2
de-
pendent (e.g. though Claim [Sh 783, 3.10=ss.6] does not apply but its proof works
using pairwise not too near

bs, in general just an uncountable set of

bs).
4) See more in 5.
Of course,
1.19 Observation. 1) For a eld K, Th(K) being strongly dependent is preserved
by nite extensions in the eld theoretic sense by 1.4(2).
2) In 1.17, if we use the -language and k
N
is innite, the theory is not strongly
dependent.
Proof. 1) Recall that by 1.11(2), the theory of the valued group (which is an ordered
abelian group) is strongly dependent, and this trivially holds for the residue eld
being nite. So by 1.15(2) we can apply part (3).
2) We consider the models of T as having three sorts: P
M
0
the eld, P
M
1
the ordered
abelian group (like value of valuations) and P
M
2
the residue eld.
Let

1
(a) I be an innite linear order, without loss of generality complete and
dense (and with no extremal members),
(b) a
t
: t I) an indiscernible sequence, a
t


C and let c C
(a singleton!)
and we shall prove

2
for some nite J I we have: if s, t IJ and (s J)(r <
I
s r <
I
t)
then a
s
, a
t
realizes the same type over c.
This suces by 2.1 and as there by 2.1(4) without loss of generality

3
a
t
= a
t,i
: i < ) list the elements of an elementary submodel M
t
of C = C
T
(we may assume M
t
is
1
-saturated; alternatively we could have assumed
that it is quite complete).
Easily it follows that it suces to prove (by the L.S.T. argument but not used)

2
for every countable u there is a nite J I which is O.K. for a
t
u :
t I).
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


22 SAHARON SHELAH
Let f
t,s
be the mapping a
s,i
a
t,i
for i < ; clearly it is an isomorphism from M
s
onto M
s
.
Now

4
p
t
= tp(c, M
t
) so (p
t
)
[]
a
for a M
t
is well dened in 1.16(a).
The case P
2
(x)

t
p
t
is easy and the case P
1
(x)

t
p
t
is easy, too, by an
assumption (and clause (c) of 1.16), so we can assume P
0
(x)

t
p
t
(x).
Let U = i < : a
s,i
P
C
0
for every ( some) s I.
Now for every i U
()
1
i
the function (s, t) val
C
(a
t,i
a
s,i
) for s <
I
t satises one of the following
Case (a)
1
i
: it is constant
Case (b)
1
i
: it depends just on s and is a strictly monotonic (increasing, by
<

) function of s
Case (c)
1
i
: it depends just on t and is a strictly monotonic (decreasing, by
<

) function of t.
[Why? This follows by inspection.]
For = 1, 0, 1 let U

:= i U : if = 0, 1, 1 then case (a)


1
i
, (b)
1
i
, (c)
1
i
respec-
tively of ()
1
i
holds so U
1
, U
0
, U
1
) is a partition of U .
For i, j U
1
we shall prove more than
()
2
i,j
we have i, j U
1
and the function (s, t) val
C
(a
t,j
a
s,i
) for s <
I
t
satises one of the following:
Case (a)
2
i,j
: val
C
(a
t,j
a
s,i
) is constant
Case (b)
2
i,j
: val
C
(a
t,j
a
s,i
) depends only on s and is a monotonic (increas-
ing) function of s and is equal to val
C
(a
s
1
,i
a
s,i
) when s <
I
s
1
Case (c)
2
i,j
: val
C
(a
t,j
a
s,i
) depends only on t and is a monotonic (increas-
ing) function of t and is equal to val
C
(a
t,j
a
t
1
,j
) when t <
I
t
1
.
[Why ()
2
i,j
holds? In this case we give full checking.
First, assume: for some (equivalently every) t I the sequence val
C
(a
t,j

a
s,i
) : s satises s <
I
t) is <

-decreasing with s recalling that we have assumed


I is a linear order with neither rst nor last element. Choose s
1
<
I
s
2
<
I
t so
by the present assumption we have val
C
(a
t,j
a
s
2
,i
) <

val
C
(a
t,j
a
s
1
,i
) hence
val
C
((a
t,j
a
s
2
,i
)(a
t,j
a
s
1
,i
)) = val
C
(a
t,j
a
s
2
,i
) which means val
C
(a
t,j
a
s
2
,i
) =
val
C
((a
s
2
,i
a
s
1
,i
)) = val
C
(a
s
2
,i
a
s
1
,i
). So in the right side t does not appear, in
the left side s
1
does not appear, hence by the equality the left side, val
C
(a
t,j
a
s
2
,i
),
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 23
does not depend on t and the right side, val
C
(a
s
2
,i
a
s
1
,i
) does not depend on s
2
but as i U
1
it does not depend on s
1
. Together by the indiscernibility for s <
I
t
we have val
C
(a
t,i
a
s,i
) is constant, i.e. case (a)
2
i,j
holds. So we can from now on
assume: for each t I the sequence val
C
(a
t,j
a
s,i
) : s satises s <
I
t) is constant
or for each t I it is <

-increasing with s.
Second, assume: for some (equivalently every) s I the sequence val
C
(a
t,j

a
s,i
) : t satises s <
I
t) is <

-decreasing with t. As in rst we can show that


case (a)
2
i,j
holds. So from now on we can assume that for every s I the sequence
val
C
(a
t,j
a
s,i
) : t satises s <
I
t) is constant or for every s I the sequence is
<

-increasing with s.
Third, assume: for some (equivalently every) t I the sequence val
C
(a
t,j
a
s,i
) :
s satises s <
I
t) is constant. This implies that s <
I
t val
C
(a
t,j
a
s,i
) = e
t
for some e = e
t
: t I). If for some (equivalently every) s I the sequence
val
C
(a
t,j
a
s,i
) : t satises s <
I
t) is constant then clearly case (a)
2
i,j
holds so we
can assume this fails so by the end of second this sequence is <

-increasing hence
e
t
: t I) is <

-increasing. So most of the requirements in case (c)


2
i,j
holds; still
we have to show that t <
I
t
1
val(a
t,j
a
t
1
,j
) = e
t
.
Let s <
I
t <
I
t
1
, we know that e
t
<

e
t
1
, which means that val
C
(a
t,j
a
s,i
) <

val
C
(a
t
1
,j
a
s,i
). This implies that val
C
((a
t,j
a
s,i
)(a
t
1
,j
a
s,i
)) = val
C
(a
t,j
a
s,i
)
which means that val
C
(a
t,j
a
t
1
,j
) = val
C
(a
t,j
a
s,i
) = e
t
as required; so case (c)
2
i,j
and we are done (if Third... holds).
Fourth, assume that for some (equivalently every) s I the sequence val
C
(a
t,j

a
s,i
) : t satises s <
I
t) is constant, then we proceed as in third getting case (b)
2
i,j
instead of case (c)
2
i,j
.
So assume that none of the above occurs, hence for every (equivalently some)
t I the sequence val
C
(a
t,j
a
s,i
) : s satises s <
I
t) is <

-increasing (with s, by
rst...and third above) and for every (equivalently some) s I the sequence
val
C
(a
t,j
a
s,i
) : t satises s <
I
t) is <

-increasing (with t, by second and


fourth above).
Hence we have s <
I
t
1
<
I
t
2
val
C
(a
t
1
,j
a
s,i
) <

val
C
(a
t
2
,j
a
s,i
)
val
C
(a
t
1
,j
a
s,i
) = val
C
((a
t
2
,j
a
s,i
) (a
t
1
,j
a
s,i
)) = val(a
t
2
,j
a
t
1
,j
) hence
val
C
(a
t
1
,j
a
s,i
) does not depend on s as s does not appear on the left side, but,
see above, it is <

-increasing with s, contradiction. So we have nished proving


()
2
i,j
.]
()
3
i
for each i U
1
, for some t

i
I + we have:
(a)
3
i
val
C
(c a
s,i
) = val
C
(a
t,i
a
s,i
) when s <
I
t and s I
<t

i
(b)
3
i
val
C
(c a
s,i
) : s I
>t

i
) is constant and if r I
>t

i
and s <
I
t are
from I
>t

i
then val
C
(c a
r,i
) <

val
C
(a
t,i
a
s,i
)
(c)
3
i
ac
C
(c a
s,i
) = ac
C
(a
t,i
a
s,i
) when s <
I
t and s I
<t

i
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


24 SAHARON SHELAH
(d)
3
i
ac
C
(c a
s,i
) : s I
>t

i
) is constant.
[Why? Recall the denition of U
1
which appeared just after ()
1
i
recalling that we
are assuming I is a complete linear order, see
1
(a).]
()
4
the set J
1
= t

i
: i U
1
has at most one member in I.
[Why? Otherwise we can nd i, j from U
1
such that t

i
,= t

j
are from I. Now apply
()
2
i,j
+ ()
3
i
+ ()
3
j
.]
So without loss of generality
()
5
J
1
is empty.
[Why? If not let J
0
= t

and we can get it is enough to prove the claim for I


<t

and for I
>t

.]
Now

1
if i U
1
and t

i
= then
(a) for every s
0
<
I
s
1
<
I
s
2
<
I
s
3
we have
(b) val
C
(x a
s
3
,i
) > val
C
(a
s
2
,i
a a
s
1
,i
) p
[]
a
s
0
,i
and
(c) c satises the formula in the left side; on p
[]
a
s
0
,j
, see
3
.
[Why? By clause (b) of 1.16 and ()
3
i
+ ()
3
j
and reect.]
Hence

2
if W
1
= i U
1
: t

i
= then
W
1
where for W U we let

W
if s <
I
t then
s,t
W
where for U

U :

s,t
U

, s, t I and f
t,s
maps p
[]
a
s,i
: i U

onto p
[]
a
t,i
: i U

.
[Why? Should be clear as J
1
= and the indiscernibility of a
t
: t I) and
1
.]

3
assume that: for every i U
1
satisfying t

i
= , there is j U
1
such
that t

j
= and s, t I val
C
(c a
t,j
) > val
C
(c a
s,i
). Then:

3
if s
0
<
I
s
1
<
I
s
2
then val
C
(x a
s
2
,j
) > val
C
(c a
s
1
,i
) p
[]
a
s
0
,i
and
the formula on the left is satised by c.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 25
[Why? Should be clear.]
Hence

4
if the assumption of
3
holds then
W
2
holds for W
2
= i U
1
: t

i
= .
[Why? As in
2
.]
Consider the assumption

5
the hypothesis of
3
fails and let j() U
1
exemplify this (so in particular
t

j()
= ). Let W
3
= i U
1
: t

i
= and val
C
(c a
s,j()
) >
val
C
(c a
t,i
) for any s, t I and W
4
= i U
1
: t

i
= and i / W
3
so
j() W
4

6
if
5
then
W
3
.
[Why? Similarly to the proof of
2
.]

7
if
5
then
(a) val
C
(c a
s,j
) : s I and j W
4
) is constant
(b) val
C
(c a
r,j()
) <

val
C
(a
t,i
a
s,i
) hence (p
s
)
[]
a
s,j()
(p
s
)
[]
a
s,i
when
i W
4
and s <
I
t r I
(c) for some nite J
1
I we have: if s, t JJ
1
and (r J
1
)(s <
I
s
r <
I
t)) then tp(val
C
(c a
s,j()
), M
s
) = f
s,t
(tp(val
C
(c a
t,j()
), M
t
))
(d) for some nite J
2
I we have: if s, t IJ
2
and (r J
r
)(r <
I
s
r <
I
t) then tp(ac
C
(c a
s,j()
), M
s
) = f
s,t
(tp(ac
C
(c a
t,j()
), M
t
)
(e) for some nite J
3
I we have: if s, t IJ
3
and (r J)(r <
I
s
r <
I
t, then
s,t
W
4
[Why? Let i W
4
; so i W
2
, hence i U
1
, which means that case (b)
1
i
of
()
1
i
holds, so for each t I the sequence val
C
(a
t,i
a
s,i
) : s satises s <
I
t)
is <

-increasing. Also as i W
2
clearly t

i
= hence by ()
3
i
(b)
3
i
we have
val
C
(ca
s,i
) : s I) is constant, call it e
i
. All this apply to j(), too. Now
as i W
4
we know that for some s
1
, t
1
I we have val
C
(c a
s
1
,j()
)

val
C
(c a
t
1
,i
), i.e. e
j()

e
i
. By the choice of j() for every j U
1
such
that t

j
= , i.e. for every j W
2
for some (equivalently every) s, t I
we have val
C
(c a
s,j
) val
C
(c a
t,j()
). In particular this holds for j = i,
hence for some s
2
, t
2
I we have val
C
(c a
s
2
,i
) val
C
(c a
t
2
,j()
), i.e
e
i

e
j()
so together with the previous sentence, e
i
= e
j()
, so clause (a)
of
7
holds. Also, the rst phrase in clause (b) is easy (using ()
3
i
(b)
3
i
second
phrase); the second phrase of (b) follows because e
i
= e
j()
. For clause (c)
note that Th(
M
) is strongly stable, for clause (d) note that Th(k
M
) is
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


26 SAHARON SHELAH
strongly dependent.
Lastly, for clause (e) combine the earlier clauses.]

8
for some nite J I, if s, t IJ and (r J)(r <
I
s r <
I
t) then
s,t
U
1
[Why? If the hypothesis of
3
holds let J = and if it fails (so
5
,
6
,
7
apply), let J be as in
7
(d), (e), so it partitions I to nitely many intervals.
It is enough to prove
s,t
W
for several W U
1
which covers U
1
. Now by
2
this holds for W
1
= i U
1
: t

i
= . If the assumption of
3
holds we
get the same for W
2
by
4
and if it fails we get it for W
3
by
6
and for W
4
by
7
(e) and the choice of J. Using U
1
= W
1
W
2
, W
2
= W
3
W
4
we are
done.]
As we can replace I by its inverse

9
for some nite J I if s, t IJ and (r)(r <
I
s r <
I
t) then
s,t
U
1
.
So we are left with U
0
. For i U
0
let e
0,i
= val(a
t,i
a
s,i
) for s <
I
t, well dened
by the denition of U
0
. Let W
5
:= i U
0
: for every (equivalently some) s ,= t I,
val
C
(c a
s,i
) < val(a
t,i
a
s,i
) and let W
6
:= U
0
W
5
.
Obviously

10
we have
W
5
.
Easily

11
if i, j W
6
then case (a)
2
i,j
of ()
2
i,j
holds.
[Why? By ()
2
i,j
and as i, j W
6
()
1
i
(a) + ()
1
i
(b).]

12
if i, j W
6
and s ,= t I then val
C
(a
t,j
a
s,i
) = e
0,i
.
[Why? As W
6
= U
0
W
5
.]
Hence

13
e
0,i
: i W
6
) is constant, call the constant value e

so s ,= t I i, j
W
6
val
C
(a
t,j
a
s,i
) = e

.
Easily

14
for every i W
6
the set I
i,c
:= s I : val
C
(c a
s,i
) > e

has at most one


member

15
let W
7
:= i W
6
: I
i,c
,= and let t

i
= I
i,c
for i W
7

16
if i, j W
7
then t

i
= t

j
.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 27
[Why? Otherwise let t I be such that t

i
< t t

j
< t, now val
C
(c a
t

i
,j
) >
val
C
(a
t,i
a
t

i
,i
) = e

and val
C
(c a
t

j
,j
) > val
C
(a
t,j
a
t

j
,j
) = e

hence e

<
val
C
((c a
t

i
,i
) (c a
t

j
,j
)) = val
C
(a
t

j
,j
a
t

i
,i
) but the last one is e

by
12
,
contradiction.]

17
without loss of generality W
7
= .
[Why? E.g. as otherwise we can prove separately for I
<t

i
and for I
>t

i
for any
i W
7
.]

18
if i, j W
6
and s ,= t I then ac
C
(ca
t,j
) ac
C
(ca
s,i
) = ac
C
(a
s,i
a
t,j
).
[Why? As val
C
(c a
t,j
), val
C
(c a
s,i
) and val
C
(c
s,i
(c
t,j
) are all equal to e

.]
The rest should be clear.
3) For the -language: the proof is similar.
1.17
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


28 SAHARON SHELAH
2 Cutting indiscernible sequence and strongly
+
dependent
2.1 Observation. 1) The following conditions on T are equivalent, for
(a) T is strongly dependent, i.e.,
0
=
ict
(T)
(b)

if I is an innite linear order, a


t


C for t I, I = a
t
: t I) is an
indiscernible sequence and C C is nite, then there is a convex equivalence
relation E on I with nitely many equivalence classes such that sEt
tp( a
s
, C) = tp( a
t
, C)
(c)

if I = a
t
: t I) is as above and C C is nite then there is a convex
equivalence relation E on I with nitely many equivalence classes such that:
if s I then a
t
: t (s/E)) is an indiscernible sequence over C.
2) We can add to the list in (1)
(b)

like (b)

but C a singleton
(c)

like (c)

but the set C is a singleton.


3) We can in part (1),(2) clauses (c)

, (b)

, (b)

, (c)

restrict ourselves to well order


I.
4) In parts (1),(2),(3), given =
<
, > [T[, in clauses (b)

, (c)

and their parallels


we can add that a

is the universe of a -saturated model; moreover we allow I


to be:
(i) I = a
u
: u [I]
<
0
) is indiscernible over A (see Denition 5.45(2))
(ii) a
{t}
= a
t
,
(iii) each a
t
is the universe of a -saturated model
(iv) for some innite linear orders I
1
, I
1
and some I

= a

u
: u [I
1
+ I +
I
1
]
<
0
) indiscernible over A = Rang( a

) we have:
() u [I]
<
0
a

u
= a
u
() for every B A of cardinality < , every subtype of the type of
a
u
: u [I
1
+ I
1
]
<
0
) over a
u
: u [I]
<
0
) of cardinality < is
realized in A (we can use only A and a
t
: t I), of course).
Remark. 1) Note that 2.8 below says more for the cases
ict
(T) >
0
so no point
to deal with it here.
2) We can in 2.1 add in (b)

, (c)

, (b

, (c

over a xed A by 1.4(3).


3) By 1.10 we can translate this to the case of a family of indiscernible sequences.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 29
Proof. 1) Let = (to serve in the proof of a subsequence observation).
(a) (b)

Let >
0
, as in the proof of 1.5, because we are assuming (a), there are
=
i
( x, y
i
) : i < ) and a
i

: i < , < ) witnessing


2

from there.
For < let a

C be the concatenation of a
i

: i < ), possibly with


repetitions so it has length .
Let = n : n < ) and

b

realizes
n
(x, a
n
n
)
n
(x, a
n
n+1
) : n < .
So for each n, tp( a
n
n
,

) ,= tp(a
n
n+1
,

) hence tp( a

n
,

) ,= tp( a

n+1
,

).
So any convex equivalence relation on as required (i.e. such that E
tp( a

) = tp( a

)) satises n < (n)E(n + 1); it certainly shows


(b)

.
(b)

(c)

Trivial.
(c)

(a)
Let a
t
: t I) and C exemplify (c)

, and assume toward contradiction that


(a) holds. Without loss of generality I is a dense linear order (so with neither rst
nor last element) and is complete and let c list C.
So
() for no convex equivalence relation E on I with nitely many equivalence
classes do we have s I a
t
: t (s/E)) is an indiscernible sequence
over C.
We now choose (E
n
, I
n
,
n
, J
n
) by induction on n such that
(a) E
n
is a convex equivalence relation on I such that each equivalence
class is dense (so with no extreme member!) or is a singleton
(b)
n
is a nite set of formulas (each of the form ( x
0
, . . . , x
m1
, y),
g( x

) = , for some m, g( y) = g( c))


(c) I
0
= I, E
0
is the equality,
0
=
(d) I
n+1
is one of the equivalence classes of E
n
and is innite
(e)
n+1
is a nite set of formulas such that a
t
: t I
n+1
) is not

n+1
-indiscernible over C
(f) E
n+1
I
n+1
is a convex equivalence relation with nitely many
classes, each dense (no extreme member) or singleton, if J
is an innite equivalence class of E
n+1
I
n+1
then a
t
: t J) is
n+1
-indiscernible over C and
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


30 SAHARON SHELAH
[I
n+1
/E
n+1
[ is minimal under those conditions
(g) E
n+1
(II
n+1
) = E
n
(II
n+1
), so E
n+1
renes E
n
(h) we choose (
n+1
, E
n+1
) such that if possible I
n+1
/E
n+1
has
4 members.
There is no problem to carry the induction as T is dependent (see 2.2(1) below
which says more or see [Sh 715, 3.4+Def 3.3]).
For n > 0, E
n
I
n
is an equivalence relation on I
n
with nitely many equivalence
classes, each convex; so as I is a complete linear order clearly
()
1
for each n > 0 there are t
n
1
<
I
. . . < t
n
k(n)1
from I
n
such that s
1
I
n
s
2

I
n
[s
1
E
n
s
2
(k)(s
1
< t
n
k
s
2
< t
n
k
s
1
> t
n
k
s
2
> t
n
k
)].
As n > 0 E
n
,= E
n1
clearly
()
2
k(n) 2 and [I
n
/E
n
[ = 2k(n) 1
()
3
I
n,
: < k(n) t
n

: 0 < < k(n) are the equivalence classes of


E
n
I
n
;
where
()
4
for non-zero n < , < k(i) we dene I
n,
:
if 0 < < k(n) 1 then I
n,
= (t
n

, t
n
+1
)
I
n
if 0 = then I
n,
= (, t
n

)
I
n
if = k(n) 1 then I
n,
= (t
n

, )
I
n
.
As (see end of clause (f))) we cannot omit any t
n

( < k(n)) and transitivity of


equality of types clearly
()
5
for each < k(n)1 for some m and = (x
0
, . . . , x
m1
, y)
n
there are
s
0
<
I
. . . <
I
s
m1
from I
n,
and s

0
<
I
. . . <
I
s

m1
from I
n,
t
n
+1
I
n,+1
such that C [= [ a
s
0
, . . . , c] [a
s

0
, . . . , c].
Hence easily
()
6
J I
n,
: < k(n) i J is a maximal open interval of I
n
such that
a
t
: t J) is
n
-indiscernible over C.
By clause (h) and ()
6
()
7
if k(n) < 4 and < k(n) then a
t
: t I
n,
) is an indiscernible sequence
over C
hence
()
8
if k(n) < 4 then for no m > n do we have I
m
I
n
.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 31
Note that
()
9
m < n I
n
I
m
I
n
I
m
= 0.
Case 1: There is an innite u such that I
n
: n < ) are pairwise disjoint.
For each n we can nd c
n

>
C and k
n
< (no connection to k(n) from above!)
and ( x
0
, . . . , x
k
n
1
, y)
n
such that a
t
: t I
n
) is not
n
( x
0
, . . . , x
k
n
1
, c)-
indiscernible (so g( x

) = ). So we can nd t

0
< . . . < t

k
n
1
in I
n
for = 1, 2 such
that [=
n
[ a
t

0
, . . . , a
t

k
n
1
, c
n
]
if(=2)
. By minor changes in
n
,
n
, without loss of generality c
n
is without repetitions hence without loss of generality n < c
n
= c

.
Without loss of generality
n
is closed under negation and without loss of
generality t
1
k
n
1
<
I
t
2
0
. We can choose t
m
k
I
n
(m < , m / 1, 2, k < k
n
)
such that for every m < , k < k
n
we have t
m
k
<
I
t
m
k+1
, t
m
k
n
1
<
I
t
m+1
0
; let
a

n,m
= a
t
m
0
. . . a
t
m
k
n
1
and let x = x
i
: i < g( c

)). So for every



the
type
n
( a

n,(n)
, x)
n
( a

n,(n)+1
, x) : n < is consistent. This is enough for
showing
ict
(T) >
0
.
Case 2: There is an innite u such that I
n
: n u) is decreasing.
For each n u, E
n
I
n
has an innite equivalence class J
n
(so J
n
I
n
) such
that n < m n, m u I
m
J
n
. By ()
8
clearly for each n u, k(n) 4
hence we can nd (n) < k(n) such that I

n
= (I
n,(n)
t
n
,n
I
n,(n)+1
) is disjoint
to J
m
. Now I

n
: n u) are pairwise disjoint and we continue as in Case 1.
By Ramsey theorem at least one of the two cases occurs so we are done.
2) By induction on [C[.
3),4) Easy by now.
2.1
Recall
2.2 Observation. 1) Assume that T is dependent, a
t
: t I) is an indiscernible
sequence, a nite set of formulas, C C nite. Then for some convex equivalence
relation E on I with nitely many equivalence classes, each equivalence class in an
innite open convex set or is a singleton such that for every s I, a
t
: t s/E) is
an -indiscernible sequence over a
t
: t I(s/E) C.
2) If I is dense and complete there is the least ne such E. In fact for J an open
convex subset of I we have: J is an E-equivalence class i J is a maximal open
convex subset of I such that a
t
: t J) is -indiscernible over C a
t
: t IJ.
3) Assume if I is dense (no extreme elements) and complete then there are t
1
<
I
. . . < t
k1
such that stipulating t
0
= , t
k
= , I

= (t

, t
+1
)
I
, we have
(a) a
t
: t I

) is indiscernible over C
(b) if 1, . . . , k 1) and t

<
I
t

<
I
t
+

, then a
t
: t (t

, t
+

)
I
) is not
-indiscernible over C.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


32 SAHARON SHELAH
Proof. 1) See clause (b) of [Sh 715, Claim 3.2].
2),3) Done inside the proof of 2.1 and see proof of 2.10.
2.2
2.3 Denition. 1) We say that =
i
( x, y
i
) : i < ) witness <
ict,2
(T) when
there are a sequence a
i,
: < , i < ) and

b
i
: i < ) such that
(a) a
i,
: < ) is an indiscernible sequence over a
j,
: j i and
< for each i <
(b)

b
i
a
j,
: j < i, <
(c) p =
i
( x, a
i,0

b
i
),
i
( x, a
i,1

b
i
) : i < is consistent (= nitely satisable
in C).
2)
ict,2
(T) is the rst such that there is no witness for <
ict,2
(T).
3) T is strongly
2
dependent (or strongly
+
dependent) if
ict,2
(T) =
0
.
4) T is strongly
2
stable if it is strongly
2
dependent and stable.
2.4 Observation. If M is a valued eld in the sense of Denition and [
M
[ > 1 then
T := Th(M) is not strongly
2
dependent.
Proof. Let a
M
be positive,
0
(x, a) := (val(x) a), E(x, y, a) := ( val(x, y)
2a) and F(x, y) = x
2
+ y (squaring in K
M
). Now for b
0
(M, a), the funciton
F(, b) is ( 2)-to-1 function from
0
(M, a) to b/E. So we can apply [Sh 783, 4].
Alternatively let a
n

M
, a
n
<

M a
n+1
for n < be such that there are
b
n,
K
M
for < such that < < val
M
(b
n,
b
n,
) > a
n
a
n
and
val(b
n
, ). Without loss of generality for eac n < the sequence b
n,
: < ) is
indiscernible over b
n
1
,
1
: n
1
n, < a
n
1
: n
1
< . Now for

clearly p

= val(x a
m,(m)
: m < n) > a
n
: n < , it is consistent, and we
have an example.
2.4
Note that the denition of strongly
2
dependent here (in 2.3) is equivalent to the
one in [Sh 783, 3.7](1) by (a) (e) of Claim 2.9 below.
The following example shows that there is a dierence even among the stable T.
2.5 Example: There is a strongly
1
stable not strongly
2
stable T (see Denition 2.3).
Proof. Fix large enough. Let F be a eld, let V be a vector space over F of
innite dimension, let V
n
: n < ) be a decreasing sequence of subspaces of V with
V
n
/V
n+1
having innite dimension and V
0
= V and V

= V
n
: n < have
dimension . Let x
n

+ V
n+1
: < ) be a basis of V
n
/V
n+1
and let x
,i

: i Z
and < ) be a basis of V

. Let M = M

be the following model:


(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 33
(a) universe: V
(b) individual constants: 0
V
(c) the vector space operations: x +y, x y and cx for c F
(d) functions: F
M
1
, a linear unary function: F
M
1
(x
n

) = x
n+1

, F
M
1
(x
,i

) =
x
,i+1

(e) F
M
2
, a linear unary function:
F
M
2
(x
0

) = x
0

, F
M
2
(x
n+1

) = x
n

and F
M
2
(x
,i

) = x
,i1

(f) predicates: P
M
n
= V
n
so P
n
unary
Now
()
0
for any models M
1
, M
2
of Th(M

) with uncountable P
M

n
: n < for
= 1, 2, the set F exemplify M
1
, M
2
are '
,
0
-equivalent where:
F is the family of partial isomorphisms f from M
1
into M
2
such that
for some n, N
i
: > 0, i = ) we have:
(a) Dom(f) =

i<n
N
i
N

(b) N
i
P
M
1
i
is a subspace when i < n i =
(c) N
i
is of nite dimension
(d) N
i
P
M
1
i+1
= 0 for i < n
(e) similar conditions on N

i
= f(N
i
) for i < n i =
()
1
T = Th(M

) has elimination of quantiers


[Why? Easy.]
Hence
()
2
T does not depend on
()
3
T is stable.
[Why? As if N
1
is
1
-saturated, N
1
N
2
then tp(a, N
1
, N
2
) : a C has
cardinality |N
1
|

0
by ()
2
.]
Now
()
4
T is not strongly
2
dependent.
[Why? By 0.1. Alternatively, dene a term
n
(y) by induction on n :

0
(y) = y,
n+1
(y) = F
1
(
n
(y)), and for

increasing let
p

(y) = P
1
(y
0
(x
0
(0))
), P
2
(y
0
(x
0
(0)
)
1
(x
1
(1)
)), . . . ,
P
n
(y

(x

()
) : < n), . . . .
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


34 SAHARON SHELAH
Clearly each p

is nitely satisable in M

. Easily this proves that T is not


strongly
2
stable I.]
So it remains to prove
()
5
T is strongly stable.
Why this holds? We work in C = C
T
. Let (2

)
+
be large enough and =

0
.
We shall prove
ict
(T) =
0
by the variant of (b)

from 2.1(3), this suces. Let


a

: < ) be an indiscernible sequence over a set A such that g( a

) . By 1.10
without loss of generality each a

enumerate the set of elements of an elementary


submodel N

of C which include A and is


1
-saturated.
Without loss of generality (I Z = and):

1
for some a

n
(n Z), A c(A

i
: i Z) and a

n
: n < 0) a

: <
) a

n
: n 0) is an indiscernible sequence over A

and a

: < )A) is
linearly independent over A

, A is the universe of N, N is
1
-saturated and
N N

is
1
-saturated (and does not depend on )
Hence by ()
0

2
(a) ,= a
,i
= a
,j
a
,i
= a
,i
A
(b) if u then cl( a

: u A) is C
(c) if u is nite we get an
1
-saturated model (not really used).
(We can use the stronger 2.1(4)). Easily

3
if a N

, b c(N

: < A) then:
(a) a = b a A
(b) a b P
C
n
(c A)(a c P
C
n
b c P
C
n
).
[Why? Let b =
C
( a

0
, . . . , a

m1
, a), a
>
A, a term,
0
<
1
< . . . <
m1
, c
>
A, then for every k < large enough b

:=
C
(a

k
, a

k+1
, . . . , a
k+m1
, a) belongs
to A and in Case (a): a = b a = b

and in case (b): a b P


C
n
a b

P
C
n
.]

4
if a

c(N

: u

A) and u

for = 1, 2 then:
(a) if a
1
= a
2
then for some b c(N

: u
1
u
2
A) we have
a
1
b = a
2
b A
(b) if a
1
a
2
P
C
n
then for some b c(N

: u
1
u
2
A) and
c A we have a
2
b c P
C
n
and a
2
b c P
C
n
.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 35
[Why? Similarly to
2
.]
Now let c C, the proof splits to cases.
Case 1: c c( a

: < A).
So for some nite u , c c( a

: u), easily a

: u) is an
indiscernible set over A c, and we are done.
Case 2: For some nite u , for every n for some c
n
c( a

: u A) we
have c c
n
P
M
n
(but not case 1).
Clearly u is as required. (In fact, easily c( a

: u A) is
1
-saturated
(as u is nite, by
2
(c)) hence there is c

c(a

: u A) such that
n < c

c
n
P
M
n
.
Case 3: Neither case 1 nor case 2 (less is needed).
Let n(1) < be maximal such that for some c
n(1)
A we have c c
n(1)
P
M
n(1)
(for n = 0 every c

A is O.K.; by not Case 2 such n(1) exists).


Subcase 3A: There is n(2) (n(1), ) and c
n(2)
c( a

: < A) such that


c c
n(2)
P
M
n(2)
.
Let u be a nite subset of such that c
n(2)
c( a

: u A), now u is as
required (by
3
+
4
above).
Subcase 3B: Not subcase 3A.
Choose u = works because neither case 1, nor case 2 hold with u = .
2.5
2.6 Remark. We can prove a claim parallel to 1.11, i.e. replacing strong dependent
by strongly
2
dependent.
2.7 Claim. 1)
ict,2
(T
eq
) =
ict,2
(T).
2) If T

= Th(M

) for = 1, 2 then
ict,2
(T
1
)
ict,2
(T
2
) when:
() M
1
is (rst order) interpretable in M
2
.
3) If T

= Th(C, c)
cA
then
ict,2
(T

) =
ict,2
(T).
4) If M is the disjoint sum of M
1
, M
2
(or the product) and Th(M
1
), Th(M
2
) are
strongly
2
dependent then so is Th(M).
Proof. Similar to 1.11.
2.7
Now
ict
(T) is very close to being equal to
ict,2
(T).
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


36 SAHARON SHELAH
2.8 Claim. 1) If =
ict,2
(T) ,=
ict
(T) then:
(a)
ict,2
(T) =
1

ict
(T) =
0
(b) there is an indiscernible sequence a
t
: t I) with a
t


C and c C, I is
dense complete for clarity such that
() for no nite u I do we have: if J is a convex subset of I disjoint to
u then a
t
: t J) is indiscernible over a
t
: t IJ c.
2) If T is strongly
+
dependent then T is strongly dependent.
3) In the denition of
ict,2
(T), without loss of generality m = 1.
Proof. 1) We use Observation 1.5. Obviously
ict
(T)
ict,2
(T), the rest is proved
together with 2.10 below.
2) Easy.
3) Similar to the proof of 1.7 or better to use 2.10(1),(2).
2.8
2.9 Claim. The following conditions on T are equivalent:
(a)
ict,2
(T) >
0
(b) we can nd A and an indiscernible sequence a
t
: t I) over A satisfying
a
t


C and t
n
I increasing with n and c
>
C such that for every n
t
n
<
I
t tp( a
t
n
, A c a
t
m
: m < n) ,= tp( a
t
, A c a
t
m
: m < n)
(c) similarly to (b) but t
n
<
I
t tp( a
t
m
, A c a
s
: s <
I
t
n
) ,= tp(a
t
, A
c a
s
: s <
I
t
n
)
(d) we can nd A and a sequence a
n
t
: t I
n
), I
n
an innite order such that
a
n
t
: t I
n
) is indiscernible over A a
m
t
: m ,= n, m < , t I
n

and for some c


>
C for each n, a
n
t
: t I
n
) is not indiscernible over
A c a
m
t
: t I
m
, m < n
(e) we can nd a sequence
n
(x, y
n
, . . . , y
0
) : n < ) and a
n

: < , n < )
such that: for every

the set p

=
n
( x, a
n

, a
n1
(n1)
, . . . , a
9
(0)
)
if(=(n))
:
n < , < is consistent.
Proof. Should be clear from the proof of 2.1 (more 2.3).
2.9
2.10 Observation. 1) For any and we have (d) (c)

(b)

(a); if
in addition we assume (
0
=
ict
(T) < =
1
=
ict,2
(T)) then we have also
(c)

(b)

so all the following conditions on T are equivalent;


(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 37
(a)
ict
(T)
(b)

if a
t
: t I) is an indiscernible sequence, I a linear order, a
t


C and
C C is nite then for some set P of < initial segments of I we have:
() if s, t I and (J P)(s J t J) then a
s
, a
t
realizes the same
type over C (if I is complete this means: for some J I of cardinality
< , if s, t I realizes the same quantier free type over J in I then
a
s
, a
t
realizes the same type over C)
(c)

like (b) but strengthening the conclusion to: if n < , s


0
<
I
. . . <
I
s
n1
, t
0
<
I
. . . <
I
t
n
and ( < n)(k < n)(J P)[s

J = t
k
J] then
a
s
0
. . . a
t
n1
and a
t
0
. . . a
t
n1
realize the same type over C
(d)
ict,2
(T).
2) We can in clause (b)

, (c)

add [C[ = 1 and/or demand I is well ordered (for the


last use 1.10).
Proof. We shall prove various implications which together obviously suce (for
2.10 and 2.8(1) and 2.8(3)).
(a) (b)

Let . As in the proof of 1.5 there are =


i
( x, y
i
) : i < ), m = g( x)
and a
i

: i < , < ) exemplifying


2

from 1.5, so necessarily a

is non-empty.
Without loss of generality g( a
0

) and without loss of generality


2
. Let
a



C be a
0

a
1

. . . a

were a

has length
<
g( a
i

) and is constantly the


rst member of a
0

. Let c realize p =
i
( x, a
2i
)
i
( x, a
2i+1
) : i < .
Easily c (or pedantically Rang( c)) and a

: < ) exemplies (b)

.
(a) (b)

.
If =
0
, this holds by 2.1(1); in general, this holds by the proof of 2.1(1) and
this is why there we use .
(b)

(c)

Obvious.
(a) (d)
The witness for (a) is a witness for (d).
(d) (c)

Let
i
( x, y
i
) : i < ) witness (d), i.e., witness <
ict,2
(T), so there are a
i,
:
< , i < ) and

b
i
: i < ) satisfying clauses (a),(b),(c) of Denition 2.3. By
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


38 SAHARON SHELAH
Observation 1.10 we can nd an indiscernible sequence a

: < ), g( a

) =

where
j
:= g( y
i
) : i < j such that i < < a

i
[
i
,
i+1
) = a
i

. Now
a

: < ), c witness (c)

, because if P is as required in (c)

then easily
(i < )(J P)(J [i, i + ] / , [i, i + , hence [P[ . Now clearly

hence repeating the rst element ( ) times we get

b
i

: < ) which
together with c exemplify (c)

.
It is enough to prove
() assume (c)

then
(i) (d)
(ii) (a) except possibly when (a) + (b) of 2.8(1) holds, in particular

0
=
ict
(T) < =
1
=
ict,2
(T).
Toward this we can assume that
T is dependent and C, a
t
: t I) form a witness to (c)

.
Let c list C without repetitions and without loss of generality I is a dense complete
linear order (so with no extreme elements). Let g( x

) = for < be pairwise


disjoint with no repetitions, of course, g( y) = g( c) < (pairwise disjoint) and let
=
i
=
i
( x
0
, . . . , x
n(i)1
, y) : i < [T[) list all such formulas in '(
T
). For each
i < [T[ by 2.2(1),(2) there are m(i) < and t
i,1
<
I
. . . <
I
t
i,m(i)1
as there and
m(i) is minimal, so stipulating t
i,0
= , t
i,m(i)
= we have:
()
1
if s

0
<
I
. . . <
I
s

m(i)1
and s

0
<
I
. . . <
I
s

m(i)1
and s

, s

realize the same


quantier free type over t
i,1
, . . . , t
i,m(i)1
in the linear order I for each
< m(i) then C [=
i
[ a
s

0
, . . . , a
s

m(i)1
, c]
i
[ a
s

0
, . . . , a
s

m(i)1
c].
For each i < [T[, for each 1, . . . , m(i) we can choose
,i
and nd w
i,
such
that
()
2
(a) w
i,
It
i,

(b) w
i,
is nite
(c) if s
1
< t
i,(i)
< s
2
then a
t
: t (s
1
, s
2
)
I
) is not
i
-indiscernible
over C a
t
: t w
i
.
If the set t
i,k
: i < [T[, k = 1, . . . , m(i) 1 has cardinality < we are done, so
assume that
()
3
t
i,
: i < [T[ and [1, m(i)] has cardinality .
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 39
Case 1: >
0
(so we have to prove (a)).
By Hajnal free subset theorem and by ()
3
there is u
0
[T[ of order type such
that i u
0
t
i,
: = 1, . . . , m(i)1 _ t
j,
: j u
0
i and = 1, . . . , m(j)1.
There are u u

of cardinality and a sequence (i) : i u), 0 < (i) < m(i)


such that t
i,(i)
: i u) is with no repetitions and disjoint to t
i,
: i u and
,= (i)

w
i,(i)
: i u. We shall now prove <
ict
(T), this gives (a), (d)
so it suces.
Clearly by 1.5 it suces to show ( any cardinality
0
)

u
there are a
i



C for i u, < and set A such that
(a) a
i

: < ) is an indiscernible sequence over a


j

: j u, j ,= i, <
A
(b) a
i

: < ) is not
i
-indiscernible over A c.
By compactness it suces to prove
v
for any nite v u and =
0
; also we can
replace by any innite linear order.
We can nd (s
1,i
, s
2,i
) : i v) such that
()
4
s
1,i
<
I
t
i,(i)
<
I
s
2,i
(for i v)
()
5
(s
1,i
, s
2,i
)
I
is disjoint to (s
1,j
, s
2,j
) : j vi

w
j,(j)
v.
So
_
a
j
t
: t (s
1,j
, s
2,j
)
I
) : j v
_
and choosing A = a
t
: t w
i,(i)
, i v are as
required above. So we are done.
Case 2: =
0
so we have to prove (d) and clause (ii) of () and (for proving
part (2) of the present 2.10) that without loss of generality [C[ = 1.
We can nd A and u

1
(a) A C
(b) u I is nite
(c) if n < and t

0
<
I
. . . <
I
t

n1
for = 1, 2 and
(k < n)(s u)(t
1
k
= s t
2
k
= s t
1
k
<
I
s t
2
k
<
I
s) then
a
t
1
0
. . . a
t
1
n1
, a
t
2
0
. . . a
t
2
n1
realize the same type over A
(d) if A

, u

satises (a)+(b)+(c) then [A

[ [A[.
This is possible because C is nite and the empty set satises clauses (a),(b),(c)
for A by our present assumption A ,= C, so let c CA. Now we try to choose
(i
k
,
k
, w
k
) by induction on k <
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


40 SAHARON SHELAH
(a) i
k
<
(b) 1
k
m(i) 1
(c) t
i
k
,
k
I
(d) w
k
u w
0
. . . w
k1
t
i
0
,k
0
, . . . , t
i
k1
,
k1

(e) w
k
It
i
k
,
k
is nite
(f) if s

<
I
t
i
k
,
k
<
I
s

then a
t
: t (s

, s

)
I
) is not indiscernible over
a
s
: s w
k
c.
If we are stuck in k then w
k1
[I]
<
0
when k > 0, u when k = 0 show that a
t
:
t I), A c contradict the choice of A and so witness (c)

. If we succeed then
we prove as in Case 1 that
ict
(Th(C, a)
aA
) >
0
so by 1.4 we get
ict
(T) >
0
.

2.10
2.11 Conclusion. T is strongly
2
dependent by Denition 2.3 i T is strongly
2
dependent by [Sh 783, 3,3.7] which means we say T is strongly
2
(or strongly
+
)
dependent when: if a
t
: t I) is an indiscernible sequence over A, t I
g( a
t
) = and

b
>
(C) then we can divide I to nitely many convex sets
I

: < k) such that for each the sequence a


t
: t I

) is an indiscernible
sequence over a
s
: s II

b.

Discussion: Now we dene T is strongly
2,
dependent, parallely to 1.8, 1.9 from
the end of 1.
2.12 Denition. 1)
icu,2
(T) is the minimal such that for no m < and =

i
( x
i
, y
i
) : i < ) with g( x
i
) = m n
i
can we nd a
i


g( y
i
)
C for < , i <
and c
,n

m
C for

such that:
(a) c
,n
: n < ) is an indiscernible sequence over a
i

: < , i <
(b) for each

and i < we have C [=
i
( c
,0
. . . c
,n
i
1
, a
i

)
if(=(i))
.
2) If is as in (1) then we say that it witnesses <
icu,2
(T).
3) T is strongly
1,
dependent if
jcu
(T) =
0
.
2.13 Claim. 1)
icu,2
(T)
ict,2
(T).
2) If cf() >
0
then
icu,2
(T) >
ict,2
(T) > .
3) The parallel of 1.4, 1.5, 1.7(2) holds.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 41
3 Ranks
(3A) Rank for strongly dependent T
3.1 Explanation/Thesis:
(a) For stable theories we normally consider not just a model M (and say a
type in it, but all its elementary extensions; we analyze them together
(b) for dependent theories we should be more liberal, allowing to replace M by
N
[a]
when M N N
1
, a
g( a)
(N
1
), (N
[ a]
is the sum of N by restrictions
of relation in N
1
denable with parameters from a
(c) this motivates some of the ranks below).
Such ranks relate to strongly
1
dependent, they have relatives for strongly
2
depen-
dent.
Note that we can represent the x K

,m
(and ranks) close to [Sh 783, 1]
particularly = 9.
3.2 Denition. 1) Let M
0

A
M
1
for M
0
, M
1
C and A C means that:
(a) M
0
M
1
(equivalently M
0
M
1
)
(b) for every

b M
1
, the type tp(

b, M
0
A) is f.g. (= nitely satisable) in
M
0
.
2) Let M
0

A,p
M
1
for M
0
, M
1
C, A C and p S
<
(M
1
A) or just p is a
(< )-type over M
1
A means that
(a) M
0
M
1
(b) if

b M
1
, c M
0
, a
1
A, a
2
A, C [=
1
[

b, a
1
, c] and
2
( x,

b, a
2
, c) p or
is just a (nite) conjunction of members of p (e.g. empty) then for some

M
0
we have C [=
1
[

1
, a
1
, c] and
2
( x,

, a
2
, c) p or just is a nite
conjunction of members of p.
3.3 Observation. 1) M
0

A,p
M
1
implies M
0

A
M
1
.
2) If p = tp(

b, M
1
A) S
m
(M
1
A) then M
0

A,p
M
1
i M
1

A

b
M
2
.
3) If M
0

A
M
1

A
M
2
then M
0

A
M
2
.
4) If M
0

A,p(M
1
A)
M
1

A,p
M
2
then M
0

A,p
M
2
.
5) If the sequences M
1,
: ), A

: ) are increasing continuous, a limit


ordinal and M
0

A

M
1,
for < then M
0

A

M
1,
. Similarly using <
A

,p

.
6) If M
1
M
2
and p is an m-type over M
1
A then M
1

A
M
2
M
1

A,p
M
2
.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


42 SAHARON SHELAH
Proof. Easy.
3.4 Discussion: 1) Note that the ranks dened below are related to [Sh 783, 1].
An alternative presentation (for 3, 6, 9, 12) is that we dene M
A
as (M, a)
aA
and T
A
= Th(C, a)
aA
and we consider p S(M
A
) and in the denition of ranks
to extend A and p we use appropriate q S(N
B
), M
A
N
A
, A B. Originally we
prsent here many variants, but now we present only two ( = 8, 9), retaining the
others in (5A).
2) We may change the denition, each time retaining from p only one formula with
little change in the claims.
3) We can dene x K
,m
such that it has also N
x
where M
x
N
x
( C
T
) and:
(A) change the denition of x

at
y to:
(a) N
y
N
x
(b) A
x
A
y
A
x
N
x
(c) M
x
M
y
N
x
(d) p
y
p
x
(B) change y explicitly

-split -strongly over x according to and replacing in
(e), (e)

p
x

by p
x
(C) dp-rk
m

,
is changed accordingly.
So now dp-rk
m

may be any ordinal so 3.7 may fail, but the result in 3 becomes
stronger covering also some model of non-strongly dependent.
3.5 Denition. 1) For = 8, 9 let
K
m,
=
_
x :x = (p, M, A), M a model C
T
, A C
T
,
p S
m
(M A) and if = 9 then
p is nitely satisable in M
_
.
If m = 1 we may omit it.
For x K
m,
let x = (p
x
, M
x
, A
x
) = (p[x], M[x], A[x]) and m = m(x) recalling p
x
is an m-type.
2) For x K
m,
let N
x
be M
x
expanded by R
( x, y, a)
=

b
g( y)
M : ( x,

b, a) p
for ( x, y, z) '(
T
), a
g( z)
A and R
( y, a)
=

b
g( y)
M : C [= [

b, a] for
( y, z) '(
T
), a
g( y)
C; let
x
=
N
x
.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 43
2A) In part (1) and (2): if we omit p we mean p = tp(<>, MA) so we can write
N
A
, a
A
-model so in this case p = (

b, a) :

b M, a M and C [= [

b, a].
3) For x, y K
m,
let
() x

pr
y means that x, y K
m,
and
(a) A
x
= A
y
(b) M
x

A[x]
M
y
(c) p
x
p
y
(d) M
x

A[x],p[y]
M
y
() x

y means that for some n and x


k
: k n), x
k

at
x
k+1
for k < n and
(x, y) = (x
0
, x
n
)
where
() x

at
y i (x, y K
m,
and) for some x

K
m,
we have
(a) x

pr
x

(b) A
x
A
y
A
x
M
x

(c) M
y
M
x

(d) p
y
p
x

(A
x
M
y
) so 1, 4 p
y
= p
x

M
y
and p
y
= p
x

(M
y
A
y
).
4) For x, y K
m,
we say that y explicitly

-splits -strongly over x when:

=
(
1
,
2
),
1
,
2
'(
T
) and for some x

and ( x, y)
2
we have clauses
(a),(b),(c),(d) of part (3)() and
(e) there are a such that
() a = a
i
: i < + 1) is
1
-indiscernible over A
x
M
y
() A
y
A
x
= a
i
: i < ; yes not + 1! (note that A
y
A
x
= and
not A
y
A
x
as we use it in (e)() in the proof of 3.7)
() a
i
M
x

for i < + 1 and



b
>
(A
x
)
() ( x, a
k

b) ( x, a

b) belongs to p
x

for k < .
5) We dene dp-rk
m

,
: K
m,
Ord by
(a) dp-rk
m

,
(x) 0 always
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


44 SAHARON SHELAH
(b) dp-rk
m

,
(x) +1 i there is y K
m,
which explicitly

-splits -strongly
over x and dp-rk
,
(y)
(c) dp-rk
m

,
(x) i dp-rk
m

,
(x) for every < when is a limit ordinal.
Clearly well dened. We may omit m from dp-rk as x determines it.
6) Let dp-rk
m

,
(T) = dp-rk
,
(x) : x K
m,
; if m = 1 we may omit it.
7) If
1
=
2
= we may write instead of (
1
,
2
). If = '(
T
) then we
may omit it.
Remark. There are obvious monotonicity and inequalities.
3.6 Observation. 1)

pr
is a partial order on K

.
2) K
m,9
K
m,8
.
3) x
8
pr
y x
9
pr
y.
4) x
8
at
y x
9
at
y.
5) y explicitly

-splits 8-strongly over x i y explicitly

-splits 9-strongly over x.
6) If x K
m,9
then dp-rk
m

,9
(x) dp-rk
m

,8
(x).
7) If a
m
C and y = (tp( a, M A), M, A) and x = (tp( a, M A), M, A) then
x K
m,8
.
8) In part (7) if tp( a, M A) is nitely satisable in M then also y K
m,9
.
9) If x K
m,
and >
0
then there is y K
m,
such that x

pr
y and
M
y
is -saturated, moreover M
y
A[y],p[y]
is -saturated (hence in Denition 3.2(4)
without loss of generality M
x

is ([M
x
A
x
[
+
)-saturated).
Proof. Easy.
3.7 Claim. 1) For each = 8, 9 we have dp-rk

(T) = i dp-rk

(T) [T[
+
i

ict
(T) >
0
.
2) For each m [1, ), similarly using dp-rk
m

(T), hence the properties do not


depend on such m.
3.8 Remark. In the implications in the proof we allow more cases of .
Proof. Part (2) has the same proof as part (1) when we recall 1.7.

ict
(T) >
0
implies dp-rk

(T) =
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 45
By the assumption there is a sequence =
n
(x, y
n
) : n < ) exemplifying

0
<
ict
(T). Let >
0
and I be Z ordered lexicographically and let I

=
Z and I

= [, ) Z. As in 1.5 by Ramsey theorem and compactness we


can nd a
n
t
: t I) (in C
T
) such that
(a) g( a
n
t
) = g( y
n
)
(b) a
n
t
: t I) is an indiscernible sequence over a
m
t
: m < , m ,= n
and t I
(c) for every

I, p

=
n
(x, a
n
t
)
if((n)=t)
: n < , t I is
consistent (i.e., nitely satisable in C).
Choose a complete T
1
T with Skolem functions and M

[= T
1
expanding C be
such that in it a
n

: t I, n < ) satises also in M

; exists by Ramsey theorem.


Let M

n
be the Skolem hull in M

of a
m
t
: m < n, t I
1
a
m
t
: m [n, ) and
t I and let M
n
= M

n
(T). So we have M
n
C which includes a
m
t
: t
I, m [n, ) such that M
n+1
M
n
and a
n
t
: t I
2
) is an indiscernible sequence
over M
n+1
a
m
t
: m < n, t I hence a
n
t
: t I
2
) is an indiscernible sequence
over M
n+1
A
m
; the indiscernibility holds even in M

where A
n
= a
m
t
: m < n
and t I
1
. We delay the case = 9. Let

I be chosen as: (2, i) : i < ). Let
p S(M
0
) be such that it includes p

.
Lastly, let x
n
= x

n
= (p
n
, M
n
, A
n
) where p
n
= p (A
n
M
n
). By 3.6(7) clearly
x
n
K

.
It is enough to show that dp-rk

(x
n
) < dp-rk

(x
n
) > dp-rk

(x
n+1
) as by
the ordinals being well ordered this implies that dp-rk

(x
n
) = for every n. By
Denition 3.5(5) clause (b) it is enough to show (xing n < ) that x
n+1
explicitly
split -strongly over x
n
using a
n
(1,i)
: i < ) a
n
(2,n)
). To show this, see Denition
3.5(4) we use x

n
:= x
n
, clearly x
n

pr
x

n
as x
n
= x

n
K

so clause (a) of Denition


3.5(3)() holds. Also A
x
n
A
x
n+1
A
x
n
M
x

n
as A
x
n+1
= A
x
n
a
n
t
: t I
1
and
a
n
t
: t I
1
M
x
n
so clause (b) of Denition 3.5(3)() holds. Also M
x
n+1
M
x

n
and p
x
n+1
p
x

n
(A
x
n
M
x
n+1
) holds trivially so also clause (c),(d) of Denition
3.5(3)() holds.
Lastly,
n
(x, a
n
(1,i)
) for i < ,
n
(x, a
(2,n)
) belongs to p

hence to p
x
n+1
hence
by renaming also clause (e) from Denition 3.5(4) holds. So we are done.
We are left with the case = 9. For the proof above to work we need just that
p( S(M
0
)) satises n < p (M
n
A
n
) is nitely satisable in M
n
. Toward this
without loss of generality for each n there is a function symbol F
n
(M

) such
that: if
n
I then c

:= F
M

n
( a
0
(0)
, . . . , a
n1
(n1)
) realizes
m
(x, a
m
t
)
if(t=(m))
:
m < n and < , so F
n
has arity g(y
m
) : m < n.
Let D be a uniform ultralter on and let c

C realize p

= (x,

b) :

b
M
0
, (x, y) '(
M
) and n : C [= (c
n
,

b) D, so clearly p = tp(c

, M
0
, C)
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


46 SAHARON SHELAH
S(M
0
) extends
n
(x, a
m
t
)
if(t=(n))
: n < and t I. So we have just to check
that p
n
= p (A
n
M
n
) is nitely satisable in M
n
, so let ( x,

b) p
n
, so we
can nd k() < ( Z) such that

b is included in the Skolem hull M

n,k()
of
a
m
(1,a)
: m < n and a Z a < k() a
m
t
: m [n, ), t I inside M

.
Let

be dened by
(m) = (m) for m [n, )
(m) = (1, k() n +m) for m < n.
By the indiscernibility:
()
1
for every n, C [= (c
n
,

b) (c
n
,

b)
and by the choice of p
()
2
n : C [= (c
n
,

b) is innite but clearly


()
3
c
m
M
n
for m < .
Together we are done.
dp-rk

(T) = implies dp-rk

(T) [T[
+
:
Trivial.
dp-rk

(T) [T[
+

ict
(T) >
0
:
We choose by induction on n sequences
n
and x
n

: < [T[
+
), a
n

: < [T[
+
)
such that:

n
(a)
n
=
m
(x, y
m
) : m < n); that is
n
=
n
m
(x, y
n
m
) : m < n) and

n
m
(x, y
n
m
) =
n+1
m
(x, y
m+1
m
) for m < n so we call it
m
(x, y
m
)
(b) x
n

and dp-rk

(x
n

)
(c) a
n

= a
n,m
,k
: k < , m < n) where the sequence a
n,m
,k
is from A
x
n

.
(d) for each < [T[
+
and m < n the sequence a
n,m
,k
: k < ) is
indiscernible over a
n,i
,k
: i < n, i ,= m, k < M
x
n

A
n

(e) we have

b
n,m

A
x
n

= a
n,i
,k
: i < m, k < A
n

for m < n such that:


if
n
and m < n

b
n,m

a
n,i
,k
: i < m, k < (i) A
n

then (p
x
n

M
x
n

)
m
( a
n,m
,(m)
,

b
n,m

)
m
( x, a
n,m
,(m)+1
,

b
n,m

) :
m < n is nitely satisable in C.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 47
For n = 0 this is trivial by the assumption rk-dp

(T) [T[
+
see Denition 3.5(6)
(and 3.5(7)).
For n + 1, for every < [T[
+
, (as rk-dp

(x
n
+1
) > by Denition 3.5(5)) we
can nd z
n

, y
n

,
n

(x, y
n

), a
n,
,k
: k < ) such that Denition 3.5(4) is satised with
(x
n
+1
, z
n

, y
n

,
n

(x, y

), a
n,
,k
: k < )) here standing for (x, x

, y, (x, y), a
k
: k <
)) there such that rk-dp

(y
n

) and we also have a


n,
,
,

b
n,

here standing for


a

b there. So for some formula


n
(x, y
n
) the set S
n
= < [T[
+
:
n

(x, y
n

) =

n
(x, y
n
) is unbounded in [T[
+
, so
n+1
is well dened so clause (a) of
n+1
holds.
For < [T[
+
let
n
() = Min(S
n
) and let x
n+1

= y
n
()
so clause (b) of
n+1
holds. Let a
n+1,m
,k
: k < ) be a
n,m
()+1,k
: k < ) if m < n and a
n,
(),k
: k < )
if m = n and let A
n+1

= A
n
()
, so clauses (c) + (d) from
n+1
holds. Also we let

b
n+1,m

is

b
n,m
()
if m < n and is

b
n,
()
if m = n. Next we check clause (e) of
n+1
.
Let
n+1
be as required in sub-clause () of clause (e) of
n+1
. By the in-
duction hypothesis (p
x
n
+1
M
x
n
+1
)(x, a
n,m
,(m)
),

b
n,m

)(x, a
n,m
,(m)+1
,

b
n,m

) :
m < n is nitely satisable in C.
By clause (d) of 3.5(3)() it follows that (p
z
n

M
z
n

) (x, a
n,m
+1,(m)
), b
n,m

)
(x, a
n,m
+1,(m)+1
) : m < is nitely satisable in C (i.e. we use M
x
n
+1

A[z
n

],p[z
n

]M[z
n

]
M
z
n

which suce; we use freely the indiscernibility).


Hence, by monotonicity for each k < and using other names, the set (p
z
n

(M
y
n

a
n+1,m
,k
: k (n) or k = A
n
+1
)( x, a
n+1,m
,(m)
,

b
m,n

)(x, a
n+1,m
,(m)+1
;

b
n,m

) :
m < n is nitely satisable in C.
Similarly (p
z
n

(M
y
n

)(x, a
n+1,m
,(n)
,

b
n+1,n

)(x, a
n+1,n
,
)(x, a
n+1,n
,(m)
,

b
n+1,m

)
(x, a
n+1,m
,(m+1)
,

b
n+1,m

) : m < n is nitely satisable in C.


But a
n+1,m
,
, a
n+1,n
,(n)+1
realizes the same type over a set including all the relevant
elements so we can above replace the rst ( a
n+1,n
,
) by the second ( a
n+1,n
,(m)+1
) so we
are done proving clause (e) of
n+1
.
Having carried the induction it suces to show that =
n
(x, y
n
) : n < )
exemplies that
ict
(T) >
0
; for this it suces to prove the assertion
2

from
1.5(1). By compactness it suces for each n to nd a
n,m
k
: k < ) for m < n in C
such that g( a
n,m
k
) = g( y
n
), a
n,m
k
: k < ) is indiscernible over a
n,i
k
: k < , i <
n, i ,= m for each m < n and C [= (x)[

m<n
((x, a
n,m
0
) (x, a
n,m
1
)].
We choose a
n,m
k
= a
n,m
,k()+k

b
n,m

where k() is large enough such that

b
n,m

:
m < n a
n,m
,k
: m < n and k < k() and let = 0; clearly we are done.
3.7
3.9 Observation. 1) If x K

and [T[ + [A
x
[ < |M
x
| then for some M
0
M
x
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


48 SAHARON SHELAH
we have |M
0
| = and for every y

pr
x satisfying M
0
M
y
we have dp-rk

(y) =
dp-rk

(x).
1A) If dp-rk

(x) < then it is < [T[


+
. Similarly dp-rk

(T), (with (2
|T|
)
+
this is
easier).
1B) If dp-rk
,
(x) < then it is < [
1

2
[
+
+
0
.
2) If x

pr
y then dp-rk

(x) dp-rk

(y).
3) If x

pr
y and z explicitly splits -strongly over y then z explicitly splits -strongly
over x.
4) The previous parts hold for m > 1, too.
Proof. 1) We do not need a really close look at the rank for this. First, x an
ordinal .
We can choose a vocabulary
,,m
of cardinality [A[ +[[ +[T[ such that:

1
for any set A xing a sequence a = a

: < ) listing the elements of


A, M C and p S
m
(M a

: < ), M
A,p
or more exactly M
a,p
is a

,,m
-model;
we let

2
(a) ds() = : a decreasing sequence of ordinals <
(b)

= u : u is a subset of ds() closed under initial segments and

: an ordinal
(c) for u

let
m
u
= : has the form
n
( x, y

) : u) where
x = x

: < m),

( x, y
n
) '(
T
) and

3
there are functions
,m
for m < , an ordinal, satisfying
(a) if u

, Ord and
m
u
, then
,m
(u) is a set of rst order
sentences
(b)
,m
(u) is a set of rst order sentences
(c) if x K
m,
and a = a

: < ) list A
x
then dp-rk

(x) i
Th(M
a,p[x]
)
,m
( ) is consistent for some
m
ds()
(d) if ,

are isomorphic (see below) then
,m
( ) is consistent i
,m
(

)
is;
where

4
=
n
( x, y

) : u),

=

( x, z

) : v) are isomorphic when there


is a one to one mapping function h from u onto v preserving lengths, being
initial segments and its negation such that

( x, y

) =
h()
( x, z
h()
) for
u.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 49
[Why
3
? just reects on the denition.]
Now if = dp-rk

(x) is (e.g. < [T[


+
) part (1) should be clear. In the
renaming case if [T[
+
by (1A) we are done and otherwise use the implicit
characterization of = dp-rk

(x).
1A) Now the proof is similar to the third part of the proof of 3.7(1) and is standard.
We choose by induction on n a formula
n
( x, y
n
) < [T[
+
for some decreasing
sequence

m,
of ordinals > of length n, we have


n,
(
n
) is consistent with Th(M
x
n

a[x
n

],p[x
n

]
) where Dom(
n,
) =

n,
:
n and
n,

n,

( x, y
n,

n,

) =

( x, y

) for < n.
The induction should be clear and clearly is enough.
1B) Similarly.
2) We prove by induction on the ordinal that dp-rk

(y) dp-rk

(x) .
For = 0 this is trivial and for a limit ordinal this is obvious. For successor
order let = + 1 so there is z K

which explicitly splits -strongly over y by


part (3) and the denition of dp-rk

we are done.
3) Easy as
pr

is transitive.
4) Similarly.
3.9

(3B) Ranks for strongly
+
dependent T:
We now deal with a relative of Denition 3.5 relevant for strongly
+
dependent.
3.10 Denition. 1) For 14, 15 we dene K
m,
= K
m,6
(and if m = 1 we
may omit it and

pr
=
6
pr
,

at
=
6
at
,

=
6
.
2) For x, y K
m,
we say that y explicitly

-split -strongly over x when:

=
(
1
,
2
),
1
,
2
'(
T
) and for some x

and ( x, y)
2
with g( x) = m we
have clauses (a),(b),(c),(d) of clause () of Denition 3.5(3) and
(e)

there are

b, a such that
() a = a
i
: i < ) is
1
-indiscernible over A
x
M
y
() A
y
A
x
a
i
: i <
()

b A
x
and a
i
M
x
for i <
() ( x, a
0

b) ( x, a
1

b) p
x

.
3) dp-rk
m

(T) = dp-rk

(x) + 1 : x K

.
4) If
1
= =
2
we may write instead of

and if = '(
T
) we may omit
. Lastly, if m = 1 we may omit it.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


50 SAHARON SHELAH
Similarly to 3.6.
3.11 Observation. 1) If x, y K
15
then y explicitly

-split 15-strongly over x i
y explicitly

-split 14-strongly over x.
2) If x K
m,15
then dp-rk
m

,15
(x) dp-rk
m

,14
(x).
Proof. Easy by the denition.
3.12 Claim. 1) For = 14 we have dp-rk

(T) = i dp-rk

(T) [T[
+
i

ict,2
(T) >
0
.
2) For each m [1, ) similarly using dp-rk
m

(T).
3) The parallel of 3.9 holds (for = 14, 15).
Proof. 1)
ict,2
(T) >
0
implies dp-rk

(T) = .
As in the proof of 3.7.
dp-rk

(T) = dp-rk

(T) [T[
+
for any .
Trivial.
dp-rk

(T) [T[
+

ict,2
(T).
We repeat the proof of the parallel statement in 3.7, and we choose

b but not
a
n+1,n
,
.
2) By part (1) and 2.8(3).
3) Similar proof.
3.12
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 51
4 Existence of indiscernibles
Now we arrive to our main result.
4.1 Theorem. 1) Assume
(a) 8, 9
(b) > () = dp-rk
m

(T) so () < [T[


+
(c)

= Q
2(()+1)
()
(d) a

C
T
for <
+

, g( a

) = m
(e) A C
T
, [A[ +[T[ .
Then for some u [
+

+
, the sequence a

: u) is an indiscernible sequence
over A.
2) If T is strongly dependent, then for some () < [T[
+
part (1) holds, i.e., if
clauses (c),(d),(e) from there holds then the conclusion there holds.
4.2 Remark. 0) This works for = 14, 15, 17, 18, too, see (5A).
1) A theorem in this direction is natural as small dp-rk points to denability and
if the relevent types increases with the index and are denable say over the rst
model then it follows that the sequence is indiscernible.
2) The Q
2(+1)
() is more than needed, we can use
+
()
where we dene

=
+ (2

)
+
: < by induction on .
3) We may like to have a one-model version of this theorem. This will be dealt with
elsewhere.
Proof. Clearly x K
m,
p
x
S
m
(A
x
M
x
) and we shall use clause (e) of
Denition 3.5(4).
By 3.6(6), it is enough to prove this for = 9, but the proof is somewhat simpler
for = 8, so we carry the proof for = 8 but say what more is needed for = 9.
We prove by induction on the ordinal that (note that the M

s are increasing but


not necessarily the p

s; this is not an essential point as by decreasing somewhat


the cardinals we can regain it):
()

if the sequence I = a

: <
+
) satises

below then for some u


[
+
]

+
the sequence a

: u) is an indiscernible sequence over A where


(below, the 2 is an overkill, in particular for successor of successor, but for
limit we catch our tail):
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


52 SAHARON SHELAH

there are , B,

M, p such that
(a) =

2(+1)
()
for every <
(b)

M = M

: <
+
) and M

C
T
is increasing continuous (with )
(c) M

has cardinality
(d) a


m
(M
+1
) for <
+
(e) p

= tp( a

, M

A B)
(f) B C, [B[
0
(g) x

= (p

, M

, A B) belongs to K
m,
and satises dp-rk
m

(x

) < .
Why is this enough? We apply () for the case = () so =

and we choose
M

C of cardinality by induction on <


+
such that M

is increasing
continuous, a

: < M

.
If = 8 ne; if = 9 it seemed that we have a problem with clause (g). That is
in checking x

K
n,
we have to show that p

is nitely satisable in M

. But
this is not a serious one: in this case note that for some club E of
+
, for every
E the type we have tp(a

, M

AB) is nitely satisable in M

. So letting
M

= M

, a

= a

when <
+
,

E and otp(C

) = and similarly
p

= tp( a

, M

, C) we can use (a

, M

, p

) : <
+
) so we are done.
So let us carry the induction; arriving to we let

= Q
2+
(), for <
3; note that

+1
=

and

2
= . Let be large enough and let B
(H (), , <

) be of cardinality such that C,



M, p, a, B, A belongs to B and
+ 1 B and Y B [Y [
2

|Y |
= X Y B. Let () =
B
+
so without loss of generality cf(()) satises
cf(())
> . Let

=
dp-rk(p
()
, M
()
, A B) and =
1
, hence =

+
. We try by induction on

+
+
+
to choose (N

) such that

(a)

< () is increasing with


(b) N

<
AB,p
()
M
()
is increasing continuous with
(c) N

has cardinality
(d) < a

(e) a

realizes p
()
(N

A B)
(f) if p
()
splits over N

AB then p
()
(N

+1
AB) splits over
N

A B
(g) (p

(N

A B), N

, A B) <
pr
(p
()
, M
()
, A B) and they
(have to) have the same dp-rk
(h) N

(but not used).


(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 53
Clearly we can carry the denition. Now the proof splits to two cases.
Case 1: For =
+
, p
()
does not split over N

A B.
By clause (e) of

clearly [, +
+
) tp( a

, N

A B) does not
split over N

A B and increases with . As N


+
: < ) is increasing and
a

N
+1
it follows that tp( a

, N

+ a

: [
+
, ) A B does not split
over N

+
1
A B. Hence by [Sh:c, I,2] that the sequence a

j
: j [, +
+
)) is
an indiscernible sequence over N

A B so as M
+

+
we are done.
Case 2: For =
+
, p
()
splits over N

A B.
So we can nd (x, y) '(
T
) and

b, c
g( y)
(M
()
AB) realizing the same
type over N

AB and ( x,

b), ( x, c) p
()
. So without loss of generality

b =

d, c = c

d where

d
>
(A B) and

b

, c

m()
(M
()
) for some m(). As
N

<
AB
M
()
(see clause (b) of

) clearly there is D, an ultralter on


m()
(N

)
such that Av(N

A B, D) = tp(

, N

A B) = tp( c

, N

A B).
Without loss of generality

m()
(N

) : ( x,

,

d) p
()
belongs to D,
as otherwise we can replace ,

, c

by , c

.
Let M

= (M
()
)
AB{ a
()
}
and let M
+
C be such that M
()
M
+
and
moreover (M

)
AB{ a
()
}
M
+
AB{ a
()
}
and the latter is
+
-saturated. Clearly
letting p
+

= (tp( a
()
, M
+
A B) and x
+
()
= (p
+
()
, M
+
()
, A B) we have
x
()

pr
x
+
()
. Note that < (p

(N

A B), N

, A B)
pr
x
()
.
We can nd

: < + ) such that



b


m()
(M
+
) realizes Av(N

A
B

: < , D) and without loss of generality

=

b

.
We would like to apply the induction hypothesis to

= dp-rk(x
()
), so let
(a)

=
(b) a

= a

for <
+
(c) M

= N

(d) p

= tp( a

, N

)
(e) B

= B

: < +
(f) A

= A.
We can apply the induction hypothesis to

, i.e., use ()

for some u


+
of
cardinality
+
the sequence a

: u

) is indisernible over A, hence the set


u :=

: u

has cardinality
+
and the sequence a

: u) is indiscernible
over A so we are done.
But we have to check that the demands from

holds (for
+
)

M

= M

: <

+
), p

= p

: <
+
).
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


54 SAHARON SHELAH
Clause (a): As = Q
2

+1
() clearly for every <

we have =

2(+1)
hence
=

2(+1)
.
Clause (b): By

(b),

M is increasing continuous.
Clause (c): By

(c).
Clause (d): By

(d).
Clause (e): By the choice of p

.
Clause (f): By the choice of B

.
Clause (g): Clearly x

K
m,
, but why do we have dp-rk(x

) <

? This is
equivalent to dp-rk(x

) < dp-rk(x
()
).
Recall x
()

pr
x
+
()
and x

explicitly split -strongly over x


()
, hence by the
denition of dp-rk we get dp-rk(x

) < dp-rk(x
()
).
What about the nitely satisable of p

when = 9? for some club E of


+
,
E tp( a

, N

A B

) is nitely satisability in N

.
2) By 3.7, dp-rk
m

(T) < [T[


+
for = 8, so we can apply part (1).
4.1
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 55
5 Concluding Remarks
We comment on some things here which we intend to continue elsewhere so the
various parts ((A),(B),...) are not so connected.
(A) Ranks for dependent theories:
We note some generalizations of 3, so Denition 3.5 is replaced by
5.1 Denition. 1) For = 1, 2, 3, 4, 5, 6, 8, 9, 11, 12 (but not 7,10), let
K
m,
=
_
x :x = (p, M, A), M a model C
T
, A C
T
,
if 1, 4 then p S
m
(M), if / 1, 4 then
p S
m
(M A) and if = 3, 6, 9, 12 then
p is nitely satisable in M
_
.
If m = 1 we may omit it.
For x K
m,
let x = (p
x
, M
x
, A
x
) = (p[x], M[x], A[x]) and m = m(x) recalling p
x
is an m-type.
2) For x K
m,
let N
x
be M expanded by R
( x, y, a)
=

b
g( y)
M : ( x,

b, a) p
for ( x, y, z) '(
T
), a
g( z)
A and = 1, 4 a =<> and R
( y, a)
=

b
g( y)
M : C [= [

b, a] for ( y, z) '(
T
), a
g( y)
C; let
x
=
N
x
.
2A) If we omit p we mean p = tp(<>, M A) so we can write N
A
, a
A
-model so
in this case p = (

b, a) :

b M, a M and C [= [

b, a].
3) For x, y K
m,
let
() x

pr
y means that x, y K
m,
and
(a) A
x
= A
y
(b) M
x

A[x]
M
y
(c) p
x
p
y
(d) if = 1, 2, 3, 8, 9 then M
x

A[x],p[y]
M
y
(for = 1 this follows from
clause (b))
() x

y means that for some n and x


k
: k n), x
k

at
x
k+1
for k < n and
(x, y) = (x
0
, x
n
)
where
() x

at
y i (x, y K
m,
and) for some x

K
m,
we have
(a) x

pr
x

(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


56 SAHARON SHELAH
(b) A
x
A
y
A
x
M
x

(c) M
y
M
x

(d) p
y
p
x

(A
x
M
y
) so 1, 4 p
y
= p
x

M
y
and / 1, 4
p
y
= p
x

(M
y
A
y
).
4) For x, y K
m,
we say that y explicitly

-splits -strongly over x when:

=
(
1
,
2
),
1
,
2
'(
T
) and for some x

and ( x, y)
2
we have clauses
(a),(b),(c),(d) of part (3)() and
(e) when 1, 2, 3, 4, 5, 6, in A
y
there is a
1
-indiscernible sequence a
k
:
k < ) over A
x
M
y
such that a
k

>
(M
x

) and ( x, a
0
), ( x, a
1
) p
x

and a
k
A
y
for k <
(e)

when = 8, 9, 11, 12 there are



b, a such that
() a = a
i
: i < + 1) is
1
-indiscernible over A
x
M
y
() A
y
A
x
= a
i
: i < ; yes not + 1! (note that A
x
= and not
A
y
A
x
as we use it in (e)() in the proof of 3.7)
()

b A
x
and a
i
M
x

for i < + 1
() ( x, a
k

b) ( x, a

b)) belongs
4
to p
x

for k < .
5) We dene dp-rk
m

,
: K
m,
Ord by
(a) dp-rk
m

,
(x) 0 always
(b) dp-rk
m

,
(x) +1 i there is y K
m,
which explicitly

-splits -strongly
over x and dp-rk
,
(y)
(c) dp-rk
m

,
(x) i dp-rk
m

,
(x) for every < when is a limit ordinal.
Clearly well dened. We may omit m from dp-rk as x determines it.
6) Let dp-rk
m

,
(T) = dp-rk
,
(x) : x K
m,
; if m = 1 we may omit it.
7) If
1
=
2
= we may write instead of (
1
,
2
). If = '(
T
) then we
may omit it.
8) For x K
m,
let x
[]
= (p
x
M
x
, M
x
, A
x
).
So Observation 3.6 is replaced by
4
this explains why = 7, 10 are missing
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 57
5.2 Observation. 1)

pr
is a partial order on K

.
2) K
m,(1)
K
m,(2)
when (1), (2) 1, 2, 3, 4, 5, 6, 8, 9, 11, 12 and (1) 1, 4
(2) 1, 4 and (2) 3, 6, 9, 12 (1) 3, 6, 9, 12.
2A) K
m,(1)
x
[]
: x K
m,(2)
when (1) 1, 4, (2) 1, . . . , 6, 8, 9, 11, 12.
2B) In (2A) equality holds if x((1), (2)) (1, 2), (1, 3), (4, 5), (4, 6).
3) x
(1)
pr
y x
(2)
pr
y when ((1), (2)) is as in (2) and (2) 2, 3, 8, 9 (1)
2, 3, 8, 9.
3B) x
(1)
pr
y x
[]

(1)
pr
y
[]
when the pair ((1), (2)) is as in (2B).
4) x
(1)
at
y x
(2)
at
y when ((1), (2)) are as in part (3) (hence (2)).
4B) x
(1)
at
y x
[]

(2)
at
y if ((1), (2)) are as in part (2A).
5) y explicitly

-splits (1)-strongly over x implies y explicitly

-splits (2)-strongly
over x when the pair ((1), (2)) is as in parts (2),(3) and (1) 1, 2, 3, 4, 5, 6
(2) 1, 2, 3, 4, 5, 6.
6) Assume ((1), (2)) is as in parts (2),(3),(5). If x K
m,(1)
then dp-rk
m

,(1)
(x)
dp-rk
m

,(2)
(x); i.e.,
(1), (2)) (3, 2), (2, 5), (3, 5), (6, 5), (3, 6)(9, 8), (8, 11), (9, 11), (12, 11), (9, 12).
7) Assume a
m
C and y = (tp( a, M A), M, A) and x = (tp( a, M A), M, A).
Then
(a) x
[]
= y
[]
(b) x K
m,1
K
m,4
(c) y K
m,2
K
m,5
K
m,8
K
m,11
(d) if tp( a, MA) is nitely satisable in M then also y K
m,3
K
m,6
K
m,9

K
m,12
.
8) If x K
m,(2)
then dp-rk

m
(2)
(x
[]
) dp-rk

m
(2)
(x) when the pair ((1), (2)) is
as in part (2A).
9) If x K
m,
and >
0
then there is y K
m,
such that x

pr
y and
M
y
is -saturated, moreover M
y
A[y],p[y]
is -saturated (hence in Denition 3.2(4)
without loss of generality M
x

is ([M
x
A
x
[
+
)-saturated).
5.3 Claim. In 3.7 we can allow = 1, 2, 5 (in addition to = 8, 9).
Proof. Similar but:

ict
(T) >
0
implies dp-rk

(T) = D when 1, 2, 4, 5, 8, 9, 11, 12:


(A) Let A
n
= a
m
t
: m < n, t I
2
if < 7 and if > 7, A
n
= a
m
t
: m < n
and t I
1
.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


58 SAHARON SHELAH
(B) x
n+1
explicitly split -strongly over x
n
using a
n
(2,n+i)
: i < ) if < 7 and
a
n
(1,i)
: i < ) a
n
2,n
) if > 7.
(C) Similarly in Lastly...: Lastly, if < 7,
n
(x, a
n
(1,n)
),
n
(x, a
n
(1,n+1)
) be-
longs to p
x

n
and even p
x
n+1
and if > 7,
n
(x, a
n
(1,n)
) for n < ,
n
(x, a
(2,n)
)
belongs to p

hence to p
x
n+1
hence by renaming also clause (e) or (e)

from
Denition 3.5(4) holds. So we are done.
dp-rk

(T) [T[
+

ict
(T) >
0
when = 1, 2, 3, 5, 6, 8, 9
(D) In
n
(e) we use
(E) () if 2, 3, 5, 6 and m < n, k < then
m
(x, a
n,m
,k
) p
x
n

k = 0 hence
m
(x, a
n,m
,k
) p
x
n

k ,= 0 for k < 2
() if = 1 then p
x
n


m
(x, a
n,m
,k
)
if(k=0)
: m < n, k < 2 is
consistent
() if = 8, 9 we also have

b
n,m

A
x
n

= a
n,i
,k
: i < m, k < A
n

for m < n such that: if


n
and m < n

b
n,m


a
n,i
,k
: i < m, k < (i) A
n

then
(p
x
n

M
x
n

)
m
( a
n,m
,(m)
,

b
n,m

)
m
( x, a
n,m
,(m)+1
,

b
n,m

) :
m < n is nitely satisable in C.
(F) In checking clause (e) of
n+1
Case = 1: We know that p
x
n
+1

m
(x, a
n,m
,k
)
if(k=0)
: m < n and k <
2 is consistent. As x
n
+1

pr
z
n

by clause ()(d) of Denition 3.5(3) we


know that q
n+1

:= p
z
n


m
(x, a
n,m
+1,k
)
if(k=0)
: m < n and k < 2 is
consistent. But
n
(x, a
n+1,m
,k
) =
n
(x, a
n,m
+1,k
) q
n+1

for k < 2, m < n and

n
(x, a
n+1
, m
,k
)
if(k=0)
=
n
(x, a
n,
,k
)
if(k=0)
q
n+1

and p
x
n+1

p
z
n

q
n+1

hence p
x
n+1

(x, a
n,m
,k
)
if(k=0)
: m n and k < 2 being a subset of q
n+1

is consistent, as required (this argument does not work for = 4).


Case 2: 2, 3, 5, 6.
Straight.
Case 3: 8, 9.
As before
5.4 Observation. Like 3.9 for = 1, 2, 3, 4, 5, 6, 8, 9, 11, 12.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 59
5.5 Denition. In Denition 3.10 we allow = 17, 18.
5.6 Observation. 1) If y explicitly

-split (1)-strongly over x then y explicitly

-
split (2)-strongly over x when ((1), (2)) (15, 14), (14, 17), (18, 17), (15, 18)
(, + 12) : = 2, 3, 5, 6.
2) If x K
m,(1)
then dp-rk
m

,(1)
(x) dp-rk
m

,(2)
(x) when ((1), (2)) is as above.
Proof. Easy by the denition.
5.7 Claim. 1) In 3.12(3) we allow = 17, 18.
2) dp-rk

(T) [T[
+

ict
(T)
1
we allow = 14, 15, 17, 18.
5.8 Theorem. In 4.1 we can allow
(a) 8, 9, 11, 12 and even 14, 15, 17, 18.
Proof. Similar to 4.1.
5.8
We can try to use ranks as in 3 for T which are just dependent. In this case
it is natural to revise the denition of the rank to make it more nitary, say
in Denition 3.5(4), clause (e),(e)

replace a
k
: k < ) by a nite long enough
sequence.
Meanwhile just note that
5.9 Claim. Let = 1, 2, 3, 5, 6 [and even = 14, 15, 17, 18]. For any nite
'(
T
) we have: for every nite
1
, rk

1
,,
(T) = i for every nite
1
,
rk

1
,,
(T) i some (x, y) has the independence property.
Proof. Similar proof to 3.7, 5.3.
Let a

: < ) M be indiscernible.
Let ( x, a
0
), ( x, a
1
) p exemplify p splits strongly over A

= M

:
< A B so tp( a
0
, A

) = tp( a
1
, A

). Let A
+
= A a
0
a
1
and we nd
u

: <
+
1
as required
() there is N
+
M, |N

| such that N

N M dp-rk(A, p
(N

A), N

) = dp-rk(A, p, M).
5.9
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


60 SAHARON SHELAH
5.10 Question: 1) Can such local ranks help us prove some weak versions of every
p S

(M) is denable? (Of course, the rst problem is to dene such weak
denability; see [Sh 783, 1]).
2) Does this help for indiscernible sequences?
5.11 Denition. We dene K
x
m,
and dx-rk
m

,
for x = p, c, q as follows:
(A) for x = p: as in Denition 3.5(4),(5), 5.1(4),(5)
(B) for x = c: as in Denition 3.5(4),(5), 5.1(4),(5) but we demand that in
clause (e),(e)

of part (4) that ( x,

b
n
) : n < is contradictory
(C) for x = q: as in Denition 3.5(4),(5), 5.1(4),(5) but clauses (e),(e)

of part
(4) we have a

from A
y
for < + such that
(x, a

)
if(<)
: < + p
x

and in (e

) we have a
n
from A
y
and
a
+n
from M
x

. In details:
(e) when 1, 2, 3, 4, 5, 6, in A
y
there is a
1
-indiscernible sequence a
k
: k <
) over A
x
M
y
such that a
k

>
(M
x

) for < and ( x, a


k
), ( x, a
+k
)
p
x

and a
k
, a
+k
A
y
for k <
(e)

when = 8, 9, 11, 12 there are



b, a such that
() a = a
i
: i < +) is
1
-indiscernible over A
x
M
y
() A
y
A
x
a
i
: i < +;
()

b A
x
and a
i
M
x

for i < +
() ( x, a
k

b) ( x, a

b) belongs
5
to p
x

for k < .]]


5.12 Question: Does Denition 5.11 help concerning question 5.10?
5.13 Discussion: We can immitate 3 with dc-rk or dq-rk instead of dp-rk and use
appropriate relatives of
ict
(T). But compare with 4.

(B) Minimal theories (or types):
It is natural to look for the parallel of minimal theories (see end of the introduction).
A subsequent work of E. Firstenberg and the author [FiSh:E50], using [Sh 757],
(see better [Sh:E63]) considered a generalization of uni-dimensional stable T.
The generalization says (see 5.22(1))
5
this explains why = 7, 10 are missing
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 61
5.14 Denition. 1) T is uni-dp-dimensional when: (T is a dependent theory and)
if I, J are innite non-trivial indiscernible sequences of singletons, then I, J have
nite distance, see below or I and J

does recalling J

is the inverse of J (i.e. we


invert the order).
2) (From [Sh:93]) for indiscernibles sequences I, J over A we say that they are im-
mediate A-neighbours if I+J is an indiscernible sequence or J+I is an indiscernible
sequence. They have distance n if there are I
0
, . . . , I
n
such that I = I
0
, J = I
n
and I

, I
+1
are immediate A-neighbors (so indiscernible over A) for < n. They
are neighbors
6
if they have distance n for some n.
3) If I is an innite indiscernible sequence over A then C
A
(I) = I

: I

, I have
distance < .
Discussion: Note that for Th(, <), the rst order theory of the rational order, any
two increasing innite sequences of elements are of distance 2. If we forget above
to have the or I, J

of nite distance, we shall get two classes up to the relevant


equivalence.
5.15 Problem: 1) Does uni-dp-dimensional theories have a dimension theory?
2) Can we characterize them?
3) If p S
m
(A), is there an indiscernible sequence I p(C) based on A?, i.e. such
that F(C
A
(I)) : F an automorphism of C over A has cardinality < C (equivalently
2
|T|+|A|
) as is the case for simple theories.
We can try another generalization.
5.16 Hypothesis. (till 5.23) Let be as in Denition 3.5, 5.1.
5.17 Denition. T is dp

-minimal when dp-rk

(x) 1 for every x K

, i.e. K
m,
for m = 1.
5.18 Remark. For this property, T and T
eq
may dier. Probably if we add only
nitely many sorts, the nite rank, i.e., dp-rk

(x) < n

< for every x K

is
preserved.
5.19 Observation. T is dp

-minimal when: for every innite indiscernible sequence


a
t
: t I), I complete, a
t


C and element c C there is t I as in 2.1 (i.e., a
singleton or the empty set if you like) when 12, and as in 2.9 when 14, . . . .
6
we may prefer the local version: for every nite L(
T
) and nite A

A (or A

= A) there
are I

, J

realizing the -type over A

of I, J respectively such that I

, J

are (innite) indiscernible


sequences over A

(or A) and has distance over A

.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


62 SAHARON SHELAH
Proof. Should be clear.
5.19
5.20 Claim. 1) For = 1, 2 we have T is dp

-minimal when: there are no a


i
n
:
n < ) and
i
(x, y
i
) such that
(a) for i = 1, 2, a
i
n
: n < ) is an indiscernible sequence over a
3i
n
: n <
(b) for some b C we have
[=
1
(b, a
1
0
)
2
(b, a
1
1
)
2
(b, a
2
0
)
2
(b, a
2
1
).
2) Similarly for rk-dp

(x) n(< ), i.e. if we replace 1 by n in Denition 5.17.


Proof. Straight.
5.21 Problem: 1) Are dp

-minimal theories T similar to o-minimal theories?


2) Characterize the dp

-minimal theories of elds.


3) What are the implications between dp

-minimal for the various .


4) Above also for uni-dp-dimensionality.
5.22 Claim. 1) For = 1, 2 the theory T, Th(1), the theory of real closed eld is
dp

-minimal; similarly for any o-minimal theory.


2) Th(1) is dp

-minimal for = 1, 2, similarly for any o-minimal theory.


3) For prime p, the rst order theory of the p-adic eld is dp
1
-minimal.
Proof. 1) As in [FiSh:E50].
2) Repeat the proof in [Sh 783, 3.3](6).
3) By the proof of 1.17.
5.22
5.23 Remark. If T is a theory of valued elds with elimination of eld quantier,
see Denition 1.14(1),(2), and k
C
T
is innite this fails. But, if
C
T
, k
C
T
are dp
1
-
minimal then the dp-rk for T are 2.
Another direction is:
5.24 Denition. 1) We say that a type p( x) is content minimal when:
(a) p( x) is not algebraic
(b) if q( x) extends p( x) and is not algebraic then
q( x)
=
p( x)
, see below.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 63
2)
p( x)
= ( x
0
, . . . , x
n1
) : p( x

) : < n ( x
1
, . . . , x
n
) is consistent, (see
[Sh:93]).
5.25 Question: Can we dene reasonable dimension for such types, at least for T
dependent or even strongly dependent?

(C) Local ranks for super dependent and indiscernibles:
Note that the original motivation of introducing strongly dependent in [Sh 783]
was to solve the equation: X/dependent = superstable/stable. However (the various
variants) of strongly dependent, when restricted to the family of stable theories,
gives classes which seem to me interesting but are not the class of superstable T.
So the original question remains open. Now returning to the search for super-
dependent we may consider another generalization of superstable.
5.26 Denition. 1) We dene lc-rk
m
(p, ) = lc-rk
m
0
(p, ) for types p which be-
longs to S
m

(A) for some A( C) and nite ( '(


T
)).
It is an ordinal or innity and
(a) lc-rk
m
(p, ) 0 always
(b) lc-rk
m
(p, ) = +1 i every < there are nite
1
and pairwise
distinct q
i
S
m

1
(A) extending p such that i < 1 + lc-rk
m
(q
i
, )
(c) lc-rk
m
(p, ) , a limit ordinal i lc-rk
m
(p) for every < .
2) For p S
m
(A) let
7
lc-rk
m
(p, ) be minlc-rk
m
(p, ) : '(
T
) nite.
3) Let lc-rk
m
(T, ) = lc-rk
m
(p, ) + 1 : p S
m
(A), A C.
4) If we omit we mean = [T[
++
.
5.27 Discussion: There are other variants and they are naturally connected to the
existence of indiscernibles (for subsets of
m
C, concerning subsets of
|T|
C), probably
representability is also relevant ([Sh:F705]).
5.28 Claim. 1) The following conditions on T are equivalent (for all > [T[
+
):
(a)

for every A and p S


m

(A) we have lc-rk


m
(p, ) <
(b)

for some

< [T[
+
for every A and p S
m

(A) we have lc-rk


m
(p, ) <

7
Easily, if
1

2
L(
T
) are nite and p
2
S
m

2
(A) and p
1
= p
2

1
then k-rk
m
(p
1
)
lc-rk
m
(p
2
). So lc-rk
m
(p, ) is well dened.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


64 SAHARON SHELAH
(c)

there is no increasing chain


n
: n < ) of nite subsets of '(
T
) and A
and p

:
>
) such that p

S
m

g()
(A) and p

and if

1
,=
2
are from
n
then p

1
,= p

2
(c)

0
like (c)

with p

:
>
).
2) Similarly restricting ourselves to A = [M[.
Proof. Easy.
5.28
Closely related is
5.29 Denition. 1) We dene lc
1
rk
m
(p, ) for types p S
m
(A) for A C as
an ordinal or innitely by:
(a) lc
1
rk
m
(p, ) 0 always
(b) lc
1
rk
m
(p, ) = +1 i for every < and nite '(
T
) we can
nd pairwise distinct q
i
S
m
(A) for i < 1 + such that p q
i
and
lc
1
rk
m
(q
i
, )
(c) lc
1
rk
m
(p, ) for a limit ordinal i lc
1
rk
m
(p) for every < .
2) If = Q
2
([T[)
++
we may omit it.
5.30 Claim. 1) The following conditions on T are equivalent when > Q
2
([T[)
++
(a)

for every A and p S


m
(A) we have lc
1
rk
m
(p, ) <
(b)

for some

< (2
|T|
)
+
for every A and p S
m
(A) we have lc
1
rk
m
(p, ) <

(c) for no A do we have a non-empty set P S


m
(A) such that for every p P
and nite '(
T
) for some nite
1
the set q
1
: q P and
q = p has cardinality > Q
2
([T[)
++
(d)

letting =
n
: n < ,
n
=

:

is a sequence of length n of nite
sets of formulas ( x, y), g( x) = m there is

:

) where

is a
nite set of formulas such that: for every we can nd A and p
,
:


and
g(

)
) such that:
() p
,
S
m
(A)
() if


n
,
n
and

n
)
n+1
, then p

,<>

n
=
p
,

n
for < and p

,<>

: < ) are pairwise


distinct
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 65
(e)

for some

:

) as above the set T

is consistent where is
non-empty and:
() if

=
n+1
,
n+1
and ( x, y)
n
then ( y)[

<g( y)
P(y

)
(( x
,
, y) ( x
n,n
, y))]
() if


n+1

n
and < < , then

(x, y)

( y)(

<g( y)
P(y

)
((x
,<>
: y) ( x
,<>
, y)).
2) Similarly restricting ourselves to the cases A = [M[, i.e. A is the universe of
some M C.
Proof. Similar.
5.30
5.31 Denition. 1) We dene lc
2
rk
m
(p, ), lc
3
rk
m
(p, ) like lc
0
rk
m
(p, ), lc
1

rk
m
(p, ) respectively replacing '(
T
) is nite by '(
T
) and arity() <
where.
2) arity() = the number of free variables of , arity() = suparity() :
(if we use the objects ( x) we may use arity(( x)) = g( x)).
5.32 Claim. The parallel of 5.30 for Denition 5.31.
Remark. Particularly the rank lc
3
rk
m
seems related to the existence of indis-
cernibility, i.e.
5.33 Conjecture: 1) Assume, lc

-rk
m
(T) < for some 3. We can prove (in
ZFC!) that for every cardinal for some we have ()
T
.
2) Moreover is not too large, say is Q
+1
( +[T[) (or just < Q
(2

)
+).

(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


66 SAHARON SHELAH
(D) Strongly
2
stable fields
A reasonable aim is to generalize the characterization of the superstable complete
theories of elds. Macintyre [Ma71] proved that every innite eld whose rst
order theory is
0
-stable, is algebraically closed. Cherlin [Ch78] proves that every
innite division ring whose rst order theory in superstable is commutative, i.e. is
a eld so algebraically closed. Cherlin-Shelah [ChSh 115] prove any superstable
theory Th(K), K an innite eld is the theory of algebraically closed elds (and
is true even for division rings). More generally we would like to replace stable by
dependent and/or superstable by strongly dependent or at least strongly
2
stable
(or other variant).
Of course, for strongly dependent we should allow at least the following cases (in
addition to the algebraically closed elds): the rst order theory of the real eld
(not problematic as is the only one with nite non-trivial Galois groups), the p-adic
eld for any prime p and the rst order theories covered by 1.17(2), i.e. Th(K
F
)
for such F.
So
5.34 Conjecture.
(a) if K is an innite eld and T = Th(K) is strongly
2
dependent (i.e.,

ict,2
(T) =
0
) then K is an algebraically closed eld (not strongly!!)
(b) similarly for division rings
(c) if K is an innite eld and T = Th(K) is strongly
1
dependent then K is
nite or algebraically closed or real closed or elementary equivalent to K
F
for some F as in 1.17(2) (like the p-adics) or a nite algebraic extension of
such a eld
(d) similarly to (c) for division rings.
Of course it is even better to answer 5.35(1):
5.35 Question: 1) Characterize the elds with dependent rst order theory.
2) At least strongly dependent (or another variant see (E),(F) below).
3) Suppose M is an ordered eld and T = Th(M) is dependent (or strongly
dependent). Can we characterize?
Remark. But we do not know this even for stability.
So adopting strongly dependent as our context we look what we can do.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 67
5.36 Claim. For a dependent T and group G interpreted in the monster model C
of T; for every (x, y) '(
T
) there is n

< such that if is nite a


i
: i < ) is
such that G (C, a
i
) is a subgroup of G then their intersection is the intersection
of some n

of them.
Remark. If T is stable this holds also for innite by the Baldwin-Saxl [BaSx76]
theorem.
Proof. See [Sh:F917].
5.37 Claim. If the complete theory T is strongly
2
dependent then nite kernel
implies almost surjectivity which means that if in C, G is a denable group, a
denable homomorphism from G into G with nite kernel then (G : Rang()) is
nite.
Proof. By a general result from [Sh 783, 3.8=tex.ss.4.5] quoted here as 0.1.
5.37
5.38 Claim. Being strongly

dependent is preserved under interpretation.


Proof. By 1.4, 2.7.
5.38
Hence the proof in [ChSh 115] works except the part on translating the connec-
tivity, which rely on ranks not available here.
However, if T is stable this is ne hence we deduce that we have
5.39 Conclusion. If K is an innite eld and Th(K) is strongly
2
stable then T is
algebraically closed.
5.40 Claim. Let p be a prime. T is not strongly dependent if T is the theory of
dierentially closed elds of characteristic p or T is the theory of some separably
closed elds of characteristic p which is not algebraically closed.
Proof. The second case implies the rst because if
1

1
, T
2
a complete '(
2
)-
theory which is strongly dependent then so is T
1
= T
2
'(
1
). So let M be a

1
-saturated separably closed eld of characteristic p which is not algebraically
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


68 SAHARON SHELAH
closed. Let
n
(x) = (y)(y
p
n
= x) and p

(x) =
n
(x) : n < and let xE
n
y
mean
n
(x y), so E
M
n
is an equivalent relation.
Let a

: <
1
) be an indiscernible set such that < <
1
a

1
(M).
Let
n
(x, y
0
, y
1
, . . . , y
n1
, z) = (z)[
n
(z) x = y
0
+y
p
1
+. . . +y
p
n1
n1
+z].
Now by our understanding of Th(M)
(a) if b

M for < n then M [= (x)


n
(x, b
0
, . . . , b
n1
)
(b) in M we have
n+1
(x, y
0
, . . . , y
n
)
n
(x, y
0
, . . . , y
n1
)
(c) in M we have, if
n
(b, a

0
, . . . , a

n1
)
n
(b, a

0
, . . . , a

n1
)
then

<n

.
[Why? Clause (a) holds because if b

M for < n then a = b


0
+b
p
1
+. . . +b
p
n1
n1
exemplies x. Clause (b) holds as if M [=
n+1
[a, b
0
, . . . , b
n1
, b
n
] as witnessed
by z d, then M [=
n
[a, b
0
, . . . , b
n1
] as witnessed by z d+b
p
n
n
which
n
(M)
as
n
(M) is closed under addition and d
n
(M) by d
n+1
(M)
n
(M)
and b
p
n
n

n
(M) as b
n
witnesses it. Lastly, to prove clause (c) assume that for
= 1, 2 we have d

= d
p
n


n
(M), b = a

0
+ a
p

1
+ a
p
2

2
+ . . . + a
p
n1

n1
+ d
p
n
2
and b = a

0
+ a
p

1
+ a
p
2

2
+ . . . + a
p
n1

n1
+ d
p
n
2
. We prove this by induction on
n. For n = 0 this is trivial, n = m + 1 substituting, etc., we get a

0
a

0
=
(a
p

1
a
p

1
) +. . . +(a
p
n1

n1
a
p
n1

n1
) +(d
p
n
2
d
p
n
1
)
1
(M), so by an assumption on
a

: <
1
) it follows that
0
=
0
. As there are unique p-th roots the original
equation implies a

1
+a
p

2
+. . . +a
p
n2

n2
+d
p
n
1
= a

1
+a
p

2
+. . . +a
p
n2

n2
+d
p
n
2
, and
we use the induction hypothesis.]
So together:
for every

(
1
), there is b

M such that
() M
n
(b

, a
(0)
, . . . , a
(n1)
) hence
() if n < ,
n
(
1
), ,= n then M [=
n
(b

, a
(0)
, . . . , a
(m1)
).
This suces.
5.40

(E) On strongly
3
dependent:
It is still not clear which versions of strong dependent (or stable) will be most
interesting. Another reasonable version is strongly
3
dependent and see more below.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 69
It has parallel properties and is natural. Hopefully at least some of those versions
allows us to generalize weight (see [Sh:c, V,3]); we intend to return to it elsewhere.
Meanwhile note:
5.41 Denition. 1) T is strongly
3
dependent if
ict,3
(T) =
0
(see below).
2)
ict,3
(T) is the rst such that the following
8
holds:
if is an ordinal, a



(M
+1
) for < , a

: [, )) is an indiscernible
sequence over M

for < and


1
<
2
M

1
M

2
C and c
>
C and
cf() then for some < , a

: [, )) is an indiscernible sequence over


M

c.
3) We say T is strongly

stable if T is strongly

dependent and is stable.


4) We dene
ict,3,
(T) and strongly
3,
dependent and strongly
3,
stable as in the
parallel cases (see Denition 1.8, 2.12), i.e., above we replace c by c
n
: n < )
indiscernible over M

: < .
5.42 Claim. 1) If T is strongly
+1
dependent then T is strongly

dependent for
= 1, 2.
2) T is strongly

dependent i T
eq
is; moreover
ict,
(T) =
ict,
(T
eq
).
3) If T
1
is interpretable in T
2
then
ict,
(T
1
)
ict,
(T
2
).
4) If T
2
= Th(B
M,MA
), see [Sh 783, 1] and T
1
= Th(M) then
ict,
(T
2
) =

ict,
(T
1
).
5) T is not strongly
3
dependent i we can nd =
n
( x
0
, x
1
, y
n
) : n < ), m =
g( x
0
)) and for any innite linear order I we can nd an indiscernible sequence
a
t
,

: t I,
>
I increasing), see Denition 5.45 below such that for any in-
creasing sequence

I, the set
n
( x
0
, a
s
,

b
n
)
if(s=(n))
: n < and (n1) <
I
s I if n > 0 of formulas is consistent (or use just s = (n), (n)+1 or (n)
I
s,
does not matter).
6) The parallel of parts (1)-(5) hold with strongly
3,
instead of strongly
3
. In par-
ticular, (parallel to part (5)), we have T is not strongly
3,
dependent i we can
nd =
n
( x
0
, . . . , x
k(n)
, y
n
) : n < ), m = g( x)) and for any innite linear
order I we can nd an indiscernible sequence a
t
,

b
,t
: t I,
>
I increasing),
see 5.45 such that for any increasing

I, ( x
0
, a
s
,

b
n
)
if(s=(n))
: n < and
(n 1) <
I
s if n > 0 ( x
i
0
, . . . , x
i
m1
, c) = ( x
j
0
, . . . , x
j
m1
, c) : m < , i
0
<
. . . < i
m1
< , j
0
< . . . < j
m1
< and c a
s
, b

: s I,
>
I increasing
is consistent.
Proof. 1)-4). Easy.
5),6) As in [Sh 897].
5.42
8
we may consider replacing by a linear order and ask for < cuts
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


70 SAHARON SHELAH
Recall this denition applies to stable T (i.e. Denition 5.41(3)).
5.43 Observation. The theory T is strongly
3
stable i: T is stable and we cannot
nd M
n
: n < ), c
>
C and a
n


(M
n+1
) such that:
(a) M
n
is F
a

-saturated
(b) M
n+1
is F
a

-prime over M
n
a
n
(c) tp( a
n
, M
n
) does not fork over M
0
(d) tp( c, M
n
a
n
) forks over M
n
.
Proof. Easy.
5.43
5.44 Conjecture For strongly
3
stable T we have dimension theory (including weight)
close to the one for superstable theories (as in [Sh:c, V]), we may try to deal with
it in [Sh 839]; related to 5(G) below.
(F) Representability and strongly
4
dependent:
In [Sh 897] we deal with T being fat or lean. We say a class K of models is fat
when for every ordinal there are a regular cardinal and non-isomorphic models
M, N K

which are EF
+
,
-equivalent where EF
+
,
is a strong version of the
isomorphism player has a winning strategy in a strong version of the Ehrenfuecht-
Frasse game of length . We prove there, that consistently if T is not strongly
stable and T
1
T, then PC(T
1
, T) is fat (in a work in preparation [Sh:F918] we
show that it suces to assume T is not strongly
4
-stable; see below).
In [Sh:F705], a work under preparation, we shall deal with representability. The
weakest form (for k a class of index models, e.g. linears order) is: an e.g. rst
order T is weakly k-represented when for every model M of T and say nite set
'(
T
) we can nd an index model I k and sequence a
t
: t I) of nite
sequences from M
C
(or just singletons) which is -indiscernible, i.e., see below,
such that [M[ a
t
: t I.
This is a parallel to stable and superstable when we play with essentially the
arity of the functions of k and the size of s considered. The thesis is that T is
stable i it, essentially can be represented for essentially k the class of sets and
parallel representability for k derived for order characterize versions of the class of
dependent theories. We also dene k-forking, i.e. replace linear orders other index
set.
We dene
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 71
5.45 Denition. 1) For any structure I we say that a
t
: t I) is indiscernible
(in C over A) when: g( a
t
) depends only on the quantier type of t in I and:
if n < and s = s
0
, s
1
, . . . , s
n1
),

t = t
0
, . . . , t
n1
) realize the same
quantier-free type in I then a
t
:= a
t
0
. . . a
t
n1
and a
s
= a
s
0
. . . a
s
n1
realize the same type (over A) in C.
2) We say that

b
u
: u [I]
<
0
) is indiscernible (in C) (over A) similarly:
if n < , w
0
, . . . , w
m1
0, . . . , n 1 and s = s

: < n),

t = t

:
< n) realize the same quantier-free types in I and u

= s
k
: k
w

, v

= t
k
: k w

then a
u
0
. . . a
u
n1
, a
v
0
. . . a
v
n1
realize the
same type in C (over A).
3) We may use incr(< , I) instead of [I]
<
0
where incr(

I) = incr

(I) = incr(, I) =
: is an increasing sequence of length of members of I; we can use < or
; clearly the dierence between incr(< , I) and [I]
<
0
is notational only (when
we have order).
5.46 Denition. 1) We say that the m-type p( x) does (, n)-ict divide over A (or
(, n)-ict
1
divide over A) when: there are an indiscernible sequence a
t
: t I), I
an innite linear order and s
0
<
I
t
0

I
s
1
<
I
t
1
<
I
. . .
I
s
n1
<
I
t
n1
such that

1
p( x) tp

( x a
s

, A) ,= tp

( x a
t

, A) for < n.
2) We say that the m-type p( x) does (, n)-ict
2
-divides over A when above we
replace
1
by:

2
p( x) tp

( x a
s

, a
s
k
: k < A) ,= tp

( x a
t

, a
s
k
: k < A)
for < n.
3) We say that the m-type p( x) does (, n)-ict
3
-divide over A when above ( a
t
:
t I incr(< n, I)) is indiscernible over A and we replace
1
by

3
p( x) tp

( x a
s

, a
s
0
,...,s
1

A) ,= tp

( x a
t

, a
s
0
,...,s
1

A) for < n.
4) We say that the m-type p( x) does (, n)-ict
4
-divide over A when there are
n

< and sequence a

: inc( n

, I)) indiscernible over A such that (where


comp(I) is the completion of the linear order I):
if c realizes p( x) then for no set J comp(I) with n members, the sequence
a

: inc( n

, I
+
)) is -indiscernible over A where I
+
= (I, P
t
)
tJ
and
P
t
:= s I : s < t. Note that if T is stable, we can equivalently require J I.
5) For k 1, 2, 3, 4 we say that the m-type p( x) does (, n)-ict
k
-forks over A
when for some sequence

( x, a

) : < () < ) we have


(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


72 SAHARON SHELAH
(a) p( x)

<()

( x, a
i
)
(b)

( x, a

) does (, n)-ict
k
-divide over A.
If k = 1 we may omit it, if = '(
T
) we may omit it.
6) We dene ict
k
rk
m
(p), an ordinal or , as follows (easily well dened):
ict
k
rk
m
(p) i p is an m-type and for every nite q p, nite A Dom(p)
and n < and < there is an m-type r extending q which ('(
T
), n) ict
k
-
forks over A with ict
k
-rk
m
(r) . If ict
k
-rk
m
(r) _ + 1; and we say that n
witnesses this if the demand above for this n fails. If n + 1 is the minimal witness
let n = ict
k
wg
n
(r).
7)
m
k,ict
(T) is the rst
0
such that for every p S
m
(B), B C there is a set
A B of cardinality < such that p does not ict
k
-fork over A. Omitting m means
for some m < ; note that we write
k,ict
(T) to distinguish it from Denition 2.3
of
ict,2
.
8) T is strongly
k
dependent [stable] if
k,ict
(T) =
0
[and T is stable].
9) We dene
k,ict,
(T) parallely i.e., now p( x) is the type of an indiscernible se-
quence of m-tuples and T is strongly
k,
dependent [stable] if it is dependent [stable]
and
k,ict,
(T) =
0
.
5.47 Claim. 1) For dependent T, the following conditions are equivalent:
(a)
4,ict,
(T) >
0
, see Denition 5.46(4),(7),(9)
(b) there are m, (

, n

) : < ), I, J such that


()

'(
T
) nite and n

< and n

> for <


() I is an innite linear order with increasing -sequence of members
() J = a

: inc
<
(I)) is an indiscernible sequence with a



C
() for

I an increasing sequence, for some c


m
C( < ) we have:
(i) c

: < ) is an indiscernible sequence over


a

: incr(I, < )
(ii) if J is the completion of the linear order I then for no
nite J
0
J do we have: if n < and

0
, . . . ,

n1

incr(I, < ) for = 1, 2 are such that
1
0
. . .
1
n1
and
2
0
. . .
2
n1
realizes the same quantier free type over J
0
in J then a

1
0
. . . a

1
n1
,
a

2
0
. . . a

2
n1
realize the same

-type over c

: < in C
(c) the natural rank is always < .
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 73
2) For dependent T the following conditions are equivalent
(a)
m
4,ict
(T) >
0
(b) like (b) is part (1) only c

: < ) is replaced by one m-tuple c


(c) ict
4
rk
m
( x = x) =
(d) ict
4
rk
m
( x = x) [T[
+
.
3) Similarly (just simpler) for k = 1, 2, 3 instead 4.
Proof. Straight, but see details Cohen-Shelah [CoSh:919].
5.47
5.48 Question: 1) Can we characterize the T such that the ict
k
-rk
1
rank of the
formula x = x is 1?
2) Do we have ict

-rk
m
( x = x) = i ict

-rk
1
(x = x) = , i.e. can we in part (2)
say that the properties do not depend on m? The positive answer will appear in
Cohen-Shelah [CoSh:919].
Now
5.49 Observation. 1) For k = 1, 2, 3 if p( x) does (, n)-ict
k
forks over A then p( x)
does (, n)-ict
k+1
forks over A.
2) If T is strongly
k+1
dependent/stable then T is strongly
k
dependent/stable.
3) For k 1, 2, 3, 4 if T is strongly
k
dependent/stable then T is strongly depen-
dent/stable; if T
1
is interpretable in T
2
and T
2
is strongly
k
dependent/stable then
so is T
1
.
4) Assume T is stable. If p S
m
(B) does not fork over A B then ict
k
-
rk
m
(p) = ict
k
rk
m
(p A).
Remark. Also the natural inequalities concerning itc
k
-rk
n
() follows by 5.49(1).
Proof. Straight.
5.49
5.50 Example: 1) There is a stable NDOP, NOTOP, not multi-dimensional count-
able complete theory which is not strongly
2
dependent.
2) T = Th(

1
(Z
2
), E
n
)
n<
is as above where Z
2
= Z/2Z as an additive group,
E
n
= (, ) : ,

1
(Z
2
) are such that (n) = (n).
3) As in part (1) but T is not strongly dependent.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


74 SAHARON SHELAH
Remark. This is [Sh 897, 0.2=0z.5]. It shows that the theorem there adds more
cases.
Proof. 1) By part (2).
2) So let M
0
be the additive group (

1
(Z
2
), +) where + is coordinatewise addition
and for let M

= (

1
(Z
2
), P
n
)
n<
, where P
n
=

1
(Z
2
) : (n) is
constantly zero and E
n
= (, ) : ,

1
(Z
2
) are such that (n) = (n)
and M

= (

1
(Z
2
), E
n
)
n<
. So M

, M

are bi-interpretable, so we shall use M

.
Let T = Th(M

) and let T

= Th(M

). So for a model N of T

is just an
abelian group in which every element has order 2, with distinguished subgraph P
N
n
for n < so a vector space over the eld Z
2
.
T is stable:
For n < , a model of T
n
is determined by nitely many dimensions: (P
N
k
:
P
N
k+1
) for < n (E
N
n
is interpreted as the equality), so T
n
is superstable not multi-
dimensional.
Hence T necessarily is stable.
T is strongly dependent not strongly
2
dependent:
As in 2.5, in fact it is strongly dependent.
T is not multi-dimensional:
If N is an
1
-saturated model of T then it is determined by the following dimen-
sion as vector spaces over Z
2
, for n <
()
1
P
N
n
/P
N
n+1
()
2

n<
P
N
n
.
Each corresponds to a regular type (in C
eq
T
).
T has NDOP:
Follows from uni-dimensionality.
T has NOTOP:
Assume N

C
T
is
1
-saturated, N
0
N

for 0, 1, 2 such that tp(N


1
, N
2
)
does not fork over N
0
. Let A be the subgroup of C generated by N
1
N
2
and let
N
3
= C
T
A. Easily N
3
C
T
, moreover N
3
is
1
-saturated.
By [Sh:c, XII] this suces.
3) Expand M

by Q
m
=

1
(Z
2
) : [m, m + ) is constantly zero for
m < n.
5.50
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 75
(G) strong
3
stable and primely minimal types
5.51 Hypothesis. T is stable (during 5(G)).
5.52 Denition. [T stable] We say p S

(A) is primely regular (usually < )


when: if > [T[ + [[ is a regular cardinal, the model M is -saturated, the type
tp( a, M) is parallel to p (or just a stationarization of it) and N is -prime over M+ a
and

b
>
N
>
M then tp( a, M +

b) is -isolated, equivalently
9
N is -prime over
M +

b.
5.53 Claim. 1) Denition 5.52 to equivalent to: there are , M, a, N as there.
2) We can in part (1) replace > [T[+[[ regular, -prime by cf() (T), F
a

-
prime respectively.
Proof. Straight.
5.53
Now (recalling Denition 5.41 and Observation 5.43).
5.54 Claim. [T is strongly
3
stable]
If cf()
r
(T) and M N are F
a

-saturated then for some a NM the type


tp(a, M) is primely regular.
Proof. The reader can note that by easy manipulations without loss of generality =
cf() > [T[; in fact, by this we can use tp instead of stp, etc.
Let

= minict
3
rk(tp(a, M)) : a NM and let a NM and

(x,

d

)
tp(a, M) be such that

= ict
3
rk(

(x,

d

)).
Let a NM. We try to choose N

, a

, B

by induction on < such that

(a) M N

N and a

M
(b) N

is F
a

-primary over M +a

and a
0
= a
(c) if = m+ 1 then
() N

N
m
and tp(a
m
, M +a

) is not F
a

-isolated
() N
m
is F
a

-primary over N

+a
m
() N

is F
a

-constructible over N
+1
+a
0
.
9
because N is -prime over M + a + c whenever c
>
N
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


76 SAHARON SHELAH
(d)() B

() a

() [B

[ <
() every F
a

-isolated type q S
<
(M B

) has no extension in
S
<
(M

B
m
: m ) which forks over M B

() B

is F
a

-atomic over M +b

.
Let (N

, a

) be dened i < 1 +() , clearly () 0.

1
if () < then tp(a
()
, M) is primely regular.
[Why? If not, then for some b N
()
M we have tp(a
()
, M+b) is not F
a

-isolated.
We try to choose

b

by induction on < such that


(
1.1
) ()

b
0
= b)
()

b


>
(N
()
)
() tp(

, M

: < b is F
a

-isolated
() tp(

, M

: < b, a
k
, . . . , a
()
is F
a

-isolated for
k = (), . . . , 0
() tp( a, M

: ) forks over M

: < for some


a
>
(B
()
) when > 0.
We are stuck for some () < because [B
()
[ < and let B

: < ().
Now we can nd an F
a

-saturated N

which is F
a

-constructible over M + B

and
F
a

-saturated N

which is F
a

-constructible over N

B
()
. By the choice of B

, the
model N

is F
a

-constructible also over M B


()
B

(by the same construction)


hence N

is F
a

-constructible over M +B
()
+B

.
Clearly N

is F
a

-prime over M + B
()
+ B

and N
()
is F
a

-prime over M +
B
()
+ B

(as B

N
()
, see clause () above) and B

has cardinality < . So


there is an isomorphism f from N

onto N
()
over M B
()
B. Renaming
without loss of generality f = id
N
so N

= N
()
.
Lastly, we shall show that (N

, b, B

) is a legal choice for (N


()+1
, a
()+1
, B
()+1
).
Why? The non-obvious clauses are (c)(), () and (d) of .
First, for clause (d) obviously B

[N

[, b f(N) and [B

[ < , so (d)(), (), ()


hold and clause (d)() holds by the clause (). As for (d)() assume q S
<
(MB

)
is F
a

-isolated let c
>
(N

) realize q, and let B


q
M B

be of cardinal-
ity < such that stp( c, B
q
) stp( c, M B

). Now we have stp( c, M B

)
stp( c, M B
()
B

) as otherwise we can nd c

in C realizing stp( c, B
q
) hence
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 77
stp( c, MB

) for = 1, 2 such that stp( c


1
, MB
()
B

) ,= stp( c
2
, MB
()
B

);
so for some nite a B
()
,

d M we have stp( c,

d aB

) ,= stp( c
2
,

d aB

) so
without loss of generality c
1
, c
2
are from N
()
contradicting the choice of (). Let

b list B

without repetitions, so by the induction hypothesis stp(

b c, M B
()
)
stp(

b c, M B
0
. . . B
()
) hence stp( c, M B
()


b) stp( c, M B
0
. . .
B
()

b) so by the choice of

b and the previous sentence really clause (d)() holds
for the choice of (N
()+1
, a
()+1
, B
()+1
) above.
Second, concerning clause (c)() of , by the sentence after the choices of
B

, N

above, we know that N

is F
a

-constructively over M B
()
B

so clearly
stp(N

, MB

) stp(N

, MB

B
()
) hence stp(B
()
, MB

) stp(B
()
, N

),
so easily stp(B
()
, M B

a
()
) stp(B
()
, N

).
Now B
()
B

is F
a

-atomic over M a
()
being N
()
recalling
()
(b)
holds; hence B
()
is F
a

-atomic over MB

a
()
hence by the previous sentence
B
()
is F
a

-atomic over N

+a
()
but [B
()
[ < hence it is F
a

-constructible over
N

+a
()
. As N

is F
a

-constructible over B
()
N

by its choice, (and a


()
B
()
by
()
(d)()), clearly N

is also F
a

-constructible over N

a
()
as required in
(c)().
Clause

(c)() means that N


()
= N

is F
a

-constructible over N

+ a
0
.
Now N
()
= N

is F
a

-constructible over B
()
N

and a
>
(N
()
) im-
plies stp( a, B
()
N

) stp( a, B
0
. . . B
()
N

) hence by monotonicity
stp( a, B
()
N

) stp( a, a
0
+B
()
+N

), so by the same construction, N


()
= N

is F
a

-constructible over a
0
+B
()
+N

. As B
()
N
()
, [B
()
[ < it is enough to
show that B
()
is F
a

-atomic over a
0
+N

and this is proved as in the proof of clause


(d)() above. So indeed (N

, b, B

) is a legal choice for (N


()+1
, a
()+1
, B
()+1
).
But this contradicts the choice of (), so we have nished proving
1
.]

2
if = m+ 1 < 1 +() then tp(a
m
, N

) is not orthogonal to M.
[Why? Toward contradiction assume tp(a
m
, N

)M. So we can nd A

of cardinality < such that tp(a


0
, . . . , a
m
), A

) is stationary, tp(a
0
, . . . , a
m
), N

)
does not fork over A

and tp(A

, M) does not fork over C

:= A

M and tp(A

, C

)
is stationary and a

and (recalling N

is F
a

-primary over M + a

) we have
stp(A

, C

+a

) stp(A

, M +a

); it follows that tp(M, A

) does not fork over C

.
As tp(a
m
, M + A

) is parallel to tp(a
m
, N

) and to tp(a
m
, A

) and tp(a
m
, N

)
M is assumed we get that all three types are orthogonal to M. It follows that
stp(a
m
, A

) stp(a
m
, M +A

) but recall a

so stp(a
m
, A

) stp(a
m
, M +a

).
As [A

[ < this implies that tp(a


m
, M+A

) is F
a

-isolated. But recall stp(A

, C

+
a

) = stp(A

, (A

M) +a

) stp(A

, M +a

). Together stp(a
m
+A

, C

+a

)
stp(a
m
+A

, M +a

) hence tp(a
m
, M +a

) is F
a

-isolated, contradicting

(c)().]
To complete the proof by
1
it suces to show () < , so toward contradiction
assume:
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


78 SAHARON SHELAH

3
() = .
As we are assuming
3
, we can nd N
+

: < () = ) such that

1
(a) N

N
+

(b) N

is saturated, e.g. of cardinality |N|


|T|
(c) N
+
+1
N
+

(d) tp(N
+

, N) does not fork over N

(e) (N
+

, c)
cN

N
+
+1
is saturated.
[Why? We can choose N
+

by induction on . For = 0 it is obvious and for


= m+1 we choose N

and satisfying the relevant demands in


1
on N

and then
choose N

m
satisfying the relevant demands on (N

, N
m
). Lastly, by the uniqueness
of saturated model there is an isomorphism f

from N

m
onto N
m
over N
m
and let
N

= f

(N

).]
Next for < () we can nd I

such that

2
(a) I

N
+

N
+
+1
(b) I

is independent over (N
+
+1
, M)
(i.e. c I

tp(c, N
+
+1
) does not fork over M and I is indepen-
dent over N
+
+1
)
(c) tp(N
+

, N
+
+1
I

) is almost orthogonal to M
(d) if c I

then either c

(C,

d

) or tp(c, M) is orthogonal to

(x,

d

), i.e. to every q S(M) to which

(x,

d

) belongs
(e) if q S(N
+
+1
) does not fork over M and

(x,

d

) q or q is
orthogonal to

(x,

d

) then the set c I

: c realizes q
has cardinality |N

|
(f) we let I

= I

(C,

d

).
[Why possible? As (N
+

, c)
cN
+
+1
) is saturated.]
Now for < ()

3
I

is not independent over (N


+
+1
+a, N
+
+1
).
[Why? Recall a = a
0
. Assume toward contradiction that
()
3.1
I

is independent over (N
+
+1
+a, N
+
+1
).
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 79
As by clause (b) of
2
we have tp(I

, N
+
+1
) does not fork over M, it follows that
I

is independent over (N
+
+1
+a, M). Also by ()
3.1
we know that tp(a, N
+
+1
I

)
does not fork over N
+
+1
. Also tp(a, N
+
+1
) does not fork over N
+1
(because a N
and tp(N
+
+1
, N) does not fork over N
+1
by
1
(d)), together it follows that
()
3.2
tp(a, N
+
+1
+I

) does not fork over N


+1
.
Recall that tp(N

, N
+
+1
) does not fork over N
+1
(by
1
(d) because N

N using
symmetry) and tp(a, N

N
+
+1
) does not fork over N

similarly hence tp(N

+
a, N
+
+1
) does not fork over N
+1
, hence

3.3
tp(N

, N
+
+1
+a) does not fork over N
+1
+a.
Recall N

is F
a

-constructible over N
+1
+ a (by
+1
(c)()), N

is F
a

-saturated
and tp(N
+
+1
, N

+a) does not fork over N


+1
clearly
()
3.4
N

is also F
a

-constructible over N
+
+1
+a (even by the same construction).
As tp(a, N
+
+1
+ I

) does not fork over N


+1
and N
+
+1
is F
a

-saturated, it follows
that
()
3.5
tp(N

, N
+
+1
+I

) does not fork over N


+
+1
hence over N
+1
.
But by
2
clause (c), for every

d
>
(N
+

) the type tp(

d, N
+
+1
+ I

) is almost
orthogonal to M hence recalling N

N
+

,
()
3.6
tp(N

, N
+
+1
+I

) is almost orthogonal to M (this does not depend on


3.1

3.5
so can be used later).
Hence by ()
3.5
+ ()
3.6
we have
()
3.7
tp(N

, N
+1
) is almost orthogonal to M.
But N
+1
is F
a

-saturated so this implies


()
3.8
tp(N

, N
+1
) is orthogonal to M.
But by

(b)
()
3.9
a

.
By
2
we have
()
3.10
tp(a

, N
+1
) is not orthogonal to M.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


80 SAHARON SHELAH
Together ()
3.8
+()
3.9
+()
3.10
give a contradiction, so ()
3.1
fails hence
3
holds.]
Now (recalling clause (f) of
2
)

4
I

is not independent over (N


+
+1
+a, N
+
+1
).
[Why? By
3
+ clauses (b)+(d) of
2
recalling that a

(C,

d

) by the choice
of a in the beginning of the proof of 5.54.]

5
for each n, tp(a, N
+
n
) does ('(
T
), n)-ict
3
-fork over M.
[Why? By 5.55 below with I

, N
+
n
here standing for I
n1
, N

there, clause (d)


there holding by
3
here. M, A there standing for M, M here, clause (a),(b),(c)
there holds by ()
3.6
here (recalling that ()
3.6
does not depend on
3.1

3.5
.]

> ict
3
rk(tp(a, N
+
n
)) for every n < .
[Why? By the choice of

(x,

d

), a,

in the beginning of the proof we have

= ict
3
rk(tp(a, M)) and by
5
and the denition of ict
3
rk() this follows.]

7
for each n, tp(a, N
+
n
) is not orthogonal to M.
[Why? By
2
(b) +
4
.]
Hence we can nd q S(M) such that:

8
(a) some automorphism of C over

d

maps tp(a, N
n
) to a type
parallel to q
(b) ict
3
rk(q) <

(c) q and tp(a, M) are not orthogonal


(d) if q

q, [q

[ < then q

(N) _ M
[actually clause (d) follows by (c)].
This contradicts the choice of

; so () < and so we are done.


5.54
5.55 Claim. Assume T is stable. A sucient condition for tp(a, N
n
) does (, n)
ict
3
-divide over A is:
(a) N

: n) is -increasing
(b) A M N
0
(c) I

N
+1
N

is independent over (N

, M) for < n
(d) tp(a, N

) forks over N
+1
(e) tp(N
+1
, N

+I

) is almost orthogonal to M.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 81
Proof. Left to the reader noting that I

: < n) are pairwise disjoint (by clauses


(a) +(c)) and I

: < n is independent).
5.55
5.56 Remark. 1) We may give more details on the last proof and intend to continue
the investigation of the theory of regular types (in order to get good theory of
weight) in this context somewhere else.
2) We can use essentially 5.55 to dene a variant of the rank for stable theory. So
5.55 can be written to use it and so 5.57 connect the two ranks.
5.57 Claim. Assume k 3, 4 and ict
k
-rk(T) < , see Denition 5.46(6).
If cf() [T[
+
or less and M N are -saturated then for some a, (x, a), n

we have:
(a) a NM
(b) if T is stable, the type p = tp(a, M) is primely regular
(c) a
>
M and (x, a) p
(d) (wict
k
-rk((x, a))) + (ict
k
wg((x, a))) is minimal.
Proof. We choose a,

(x,

d

), , n

such that
(a) a NM
(b)

d

M
(c) C [= [a,

d

]
(d) = ict
k
rk(

(x,

d

))
(e) under clauses (a)-(d), the ordinal is minimal
(f) n

witness + 1 _ ict
k
rk((x,

d

))
(g) under clauses (a)-(f) the number n

(< ) is minimal.
Clearly there are such a,

(x, c), and n

. Then we try to choose (N

, a

) by
induction on < such that

from the proof of 5.54 holds. But now we can


prove similarly that () n

. But still tp(a, N


()
) is not orthogonal to M.
[Why? We can choose N
+
0
, . . . , N
+
()
, I
0
, . . . , I
()1
as in
2
+
3
in the proof of 5.53
and prove
3
there which implies the statement above. As

(x,

d

) tp(a, N
()
)
it follows that (N
()
, c) _ M and any a

(N
()
, c)M is as required.]
This is enough.
5.57
Similarly to Denition 5.46.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


82 SAHARON SHELAH
5.58 Denition. Let T be stable.
1) For an m-type p( x) we dene sict
3
-rk
m
(p( x)) as an ordinal or by dening
when ict
3
-rk
m
(p( x)) for an ordinal by induction on
()

p( x)
sict
3
-rk
m
(p( x)) i for every < and nite q( x) p(x) and n < we
have
()
,n
q( x)
we can nd M

: n), I

: < n) and a
(a) M

C is F
a

1
(T)
-saturated
(b) M

M
+1
(c) q( x) is an m-type over M
0
(d) a realizes q( x) and sict
3
rk(tp( a, M
n
))
(e) I


>
(M
+1
) is independent over (M

, M
0
)
(f) I

is not independent over (M

+ a, M
0
)
(clearly without loss of generality I

is a singleton).
2) If sict
3
-rk
m
(p( x)) = < then we let sict
3
-wg
m
(p( x)) be the maximal n such
that for every nite q( x) p( x) we have ()
,n
q( x)
.
3) Above instead sict
3
-rk(tp( a, A)) we may write sict
3
-rk
m
( a, A); similarly for scit
3
-
wg
m
( a, A); if m = 1 we may omit it.
5.59 Claim. 1) T is strongly
3
stable i T is stable and sict
3
-rk
m
(p( x)) < for
every m-type p( x).
2) For every type p( x) there is a nite q( x) p( x) such that (sict
3
-rk(p( x)), sict
3
-
wg(p( x)) = sict
3
rk(q( x)), sict
3
-wg(q( x))).
3) If p( x) q( x) then sict
3
-rk(p( x)) sict
3
-rk(q( x)) and if equality holds then
sict
3
-wg
m
(p( x)) sict
3
-wg
m
(q( x)).
4) (T stable) If p( x), q( x) are stationary parallel types, then sict
3
-rk
m
(p( x)) = sict
3
-
rk
m
(q( x)), etc. If a
1
, a
1
realizes p S
m
(A) then sict
3
-rk
m
(rm stp( a
1
, A)) = sict
3
-
rk
m
(stp( a
2
, A)). Similarly for sict
3
-wg
m
. Also automorphisms of C preserve sict
3
-
rk
m
and sict
3
-wg.
5.60 Claim. p( x) does (, n)-ict
3
forks over A for every n when:
(a) G is a denable group over A (in C)
(b) b G realizes a generic type of G from S(A) as was proved to exist in
[Sh 783, 4.11], or T stable
(c) p( x) S
<
(A+b) forks over A.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 83
Remark. We may have said it in 5(F).
Proof of 5.60. Straight.
5.61 Conclusion.: Assume T is strongly
3
dependent.
If G is a type-denable group in C
T
then there is no decreasing sequence G
n
:
n < ) of subgroups of G such that (G
n
: G
n+1
) = for every n.
5.62 Remark. 1) In 5.60 we can replace ict
3
: by ict
4
and also by suitable
variants for stable theories.
2) Similarly in 5.61.
(H) T is n-dependent
On related problems and background see [Sh 702, 2.9-2.20], (but, concerning
indiscernibility, it speaks on nite tuples, i.e. < in 5.71, which aect the
denitions and the picture). On a consequence of T is 2-dependent for denable
subgroups in C (and more, e.g. concerning 5.64), see [Sh 886].
5.63 Denition. 1) A (complete rst order) theory T is n-independent when
clause (a)
n
in 5.64 below holds.
2) The negation isn-dependent.
5.64 Problem Sort out the relationships between the following candidates for T is
n-independent (T is order order complete, also we can x ; omitting m we mean
1)
(a)
n
some ( x, y
0
, y
1
, . . . , y
n1
) is n-independent, i.e. (a)
n
m
for some m
(a)
n
m
some ( x, y
0
, y
1
, . . . , y
n1
) is n-independent where g( x) = m where
( x, y
0
, y
1
, . . . , y
n1
) is n-independent when there are a


g( y

)
C for
< , < n and ( x, a
0
(0)
, . . . , a
n1
(n1)
) :
n
is increasing) is an
independent (sequence of formulas)
(b)
n
m
there is an indiscernible sequence a

: < ), = ( x, y
0
, . . . , y
n1
), m =
g( x), g( y

) = g( a

) for < n, < and c


g( x)
C such that:
if k < n and R

: < ()) is a nite sequence of k-place relations on


then for some sequence

t, s
n
realizing the same quantier free type
in (, <, R
0
, R
1
, . . . , R
()
) we have C [= [

b, a
s
0
, . . . , a
s
n1
][

b, a
t
0
, . . . , a
t
n1
]
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


84 SAHARON SHELAH
(c)
n
m
for some = ( x, y
0
, . . . , y
n1
), g( x) = m, for every j [1, ), for innitely
many k there are a

i

g( y)
C for i < k such that [p ( x, a
0
i
0
, . . . , a
n1
i
n1
) :
i

< k for < n : p S


m
( a

i
: < n, i < k[[ 2
k
n1
m
.
Remark. We can phrase (b)
n
m
, (c)
n
m
as alternative denitions of ( x, y
0
, . . . , y
n1
)
is n-independent. So in (b)
n
m
better to have n indiscernible sequences.
5.65 Observation. If ( x, y
0
, . . . , y
n1
) satises clause (a)
n
then it satises a strong
form of clause (c)
n
(for every k and the number is 2
k
n
.
Remark. Clearly Observation 5.65 can be read as a sucient condition for being
n-dependent, e.g.
5.66 Conclusion. T is n-dependent when: for every m, and nite '(
T
) for
innitely many k < we have [A[ k [S
m

(A)[ < 2
(k/)
n
.
5.67 Question: 1) Can we get clause (a) from clause (c)?
2) Can we use it to prove (a)
n
1
(a)
n
m
?
5.68 Observation. In 5.64, if clause (a) then clause (b).
5.69 Question: Does (b) imply (a)?
5.70 Claim. If T satises (a)
n
for every n then: if ()
<
2
then
T
()

0
where
5.71 Denition. We say that
T
()

when: if a
i


(C
T
) for i < then for
some U []

the sequence a
i
: i U ) is an indiscernible sequence in C
T
.
Remark. 1) Note that for < this property behaves dierently.
2) Of course, if = 2
||+|T|
and ()
<

then
T
()

.
3) See on the non-2-independent T and denable groups in [Sh 886].
5.72 Conjecture. Assume (a)
n
(or another variant of n-dependent). Then ZFC
(
T
()

).
5.73 Question: Can we phrase and prove a generalization of the type-decomposition
theorems for dependent theories ([Sh:900]) to n-dependent theories T, e.g. when
(

+1
) =
+1
for < n, B

(H (
+
), , <


+
) has cardinality

, [B
+1
]

, C
T
, B
+1
, . . . , B
n
B

.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 85
REFERENCES.
[BaSx76] John T. Baldwin and Jan Saxl. Logical stability in group theory. J.
Austral. Math. Soc. Ser. A, 21:267276, 1976.
[Ch78] Gregory Cherlin. Superstable division rings. In Logic Colloquium 77
(Proc. Conf., Wroclaw, 1977), volume 96 of Stud. Logic Foundations
Math, pages 99111, Amsterdam New York, 1978. NorthHolland.
[ChSh 115] Gregory Cherlin and Saharon Shelah. Superstable elds and groups.
Annals of Mathematical Logic, 18:227270, 1980.
[CLR06] R. Cluckers, L. Lipshitz, and Z. Robinson. Analytic cell decomposi-
tion and analytic motivic integration. Ann. Sci. Ecole Norm. Sup., to
appear. math.AG/0503722.
[CoSh:E65] Moran Cohen and Saharon Shelah. On strongly dependent theories.
[CoSh:919] Moran Cohen and Saharon Shelah. Stable theories and representation
over sets. Mathematical Logic Quarterly, submitted. 0906.3050.
[FiSh:E50] Eyal Firstenberg and Saharon Shelah. Pseudo unidimensional depen-
dent theories.
[Gu77] Yuri Gurevich. Monadic theory of order and topology. I. Israel Journal
of Mathematics, 27:299319, 1977.
[Ma71] Angus Macintyre. On
1
categorical theories of elds. Fundamenta
Mathematicae, 71:125, 1971.
[On0x1] Alf Onshuus. th-Forking, Algebraic Independence and Examples of
Rosy Theories. Preprint, 2003. math.LO/0306003.
[On0x2] Alf Onshuus. Properties and Consequences of th-Independence.
Preprint, 2004. math.LO/0205004.
[Pa90] Johan Pas. Cell decomposition and local zeta functions in a tower
of unramied extensions of a p-adic eld. Proc. London Math. Soc.,
60:3767, 1990.
[Sh:F931] Saharon Shelah. A dependent dream and recounting types.
[Sh 886] Saharon Shelah. Denable groups for dependent and 2-dependent the-
ories. Journal of Symbolic Logic, submitted. math.LO/0703045.
[Sh 877] Saharon Shelah. Dependent T and Existence of limit models. Tbilisi
Mathematical Journal, submitted. math.LO/0609636.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


86 SAHARON SHELAH
[Sh:900] Saharon Shelah. Dependent theories and the generic pair conjec-
ture. Communications in Contemporary Mathematics, submitted.
math.LO/0702292.
[Sh:906] Saharon Shelah. No limit model in inaccessibles. Proceedings of the
Makkaifest, accepted. 0705.4131.
[Sh:E63] Saharon Shelah. Quite Complete Real Closed Fields revisited.
[Sh:F705] Saharon Shelah. Representation over orders of elementary classes.
[Sh 839] Saharon Shelah. Stable Frames and weight. Preprint.
[Sh:F917] Saharon Shelah. Strongly dependent theories and denable groups.
[Sh:F918] Saharon Shelah. Theories with EF-Equivalent Non-Isomorphic Models
II.
[Sh:93] Saharon Shelah. Simple unstable theories. Annals of Mathematical
Logic, 19:177203, 1980.
[Sh 197] Saharon Shelah. Monadic logic: Hanf numbers. In Around classication
theory of models, volume 1182 of Lecture Notes in Mathematics, pages
203223. Springer, Berlin, 1986.
[Sh:c] Saharon Shelah. Classication theory and the number of nonisomorphic
models, volume 92 of Studies in Logic and the Foundations of Math-
ematics. North-Holland Publishing Co., Amsterdam, xxxiv+705 pp,
1990.
[Sh 702] Saharon Shelah. On what I do not understand (and have some-
thing to say), model theory. Mathematica Japonica, 51:329377, 2000.
math.LO/9910158.
[Sh 715] Saharon Shelah. Classication theory for elementary classes with the
dependence property - a modest beginning. Scientiae Mathematicae
Japonicae, 59, No. 2; (special issue: e9, 503544):265316, 2004.
math.LO/0009056.
[Sh 757] Saharon Shelah. Quite Complete Real Closed Fields. Israel Journal of
Mathematics, 142:261272, 2004. math.LO/0112212.
[Sh 876] Saharon Shelah. Minimal bounded index subgroup for dependent the-
ories. Proceedings of the American Mathematical Society, 136:1087
1091, 2008. math.LO/0603652.
[Sh 897] Saharon Shelah. Theories with EF-Equivalent Non-Isomorphic Models.
Tbilisi Mathematical Journal, 1:133164, 2008. math.LO/0703477.
(
8
6
3
)


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







m
o
d
i
f
i
e
d
:
2
0
0
9
-
0
9
-
2
3


STRONGLY DEPENDENT THEORIES SH863 87
[Sh 783] Saharon Shelah. Dependent rst order theories, continued. Israel Jour-
nal of Mathematic, 173:160, 2009. math.LO/0406440.