Saharon Shelah and Alexander Usvyatsov - More On SOP1 and SOP2

8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

MORE ON SOP
1
AND SOP
2
SAHARON SHELAH AND ALEXANDER USVYATSOV
Abstract. This paper continues [Sh500] and [DzSh692]. We present
a rank function for NSOP
1
theories and give an example of a theory
which is NSOP
1
but not simple. We also investigate the connection
between maximality in the ordering
among complete rst order

theories and the (N)SOP
2
property. We prove that
-maximality
implies SOP
2
and obtain certain results in the other direction. The
paper provides a step toward the classication of unstable theories
without the strict order property.
1. Introduction and preliminaries
We continue the work started by Mirna Dzamonja and the rst au-
thor in [Sh500] and [DzSh692]. The main goal of this project is to
throw more light on rst order theories with the tree property (that is,
non-simple) and without the strict order property (more specically,
without the SOP
3
, see Denition 1.1 below). We pursue and nalize
certain directions started in [DzSh692] and answer several questions
asked there, providing a more general and complete picture.
The reader may be familiar with a former version of this paper that
has been available as a preprint on Shelahs archive (under the number
Publication no. 844 on Shelahs list of publications. The authors would like to
thank the Israel Science Foundation for partial support of this research (Grant
no. 242/03). Mirna Dzamonja has the second authors warm thanks for fruitful
discussions and support. The authors thank Shani Ben-David for kindly typing
parts of this paper.
1
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

2 SAHARON SHELAH AND ALEXANDER USVYATSOV
ShUs:E32) and on Usvyatsovs webpage, and is referred to in the
most recent version of [DzSh692].
Some connections between the work of Dzamonja and Shelah and
this article are already explained in the introduction of [DzSh692]. In
particular, our results provide a generalization of the main theorem
of section 1 in [DzSh692], expand the results of section 2 there and
answer certain questions which were left open in section 3. One of the
answers leads to a complete proof of a theorem which had been the
original motivation of section 3 of [DzSh692], Corollary 3.15 here (see
also Discussion 3.12 here, Theorem 0.5 in [DzSh692] and discussion
preceding it). We give more details below.
Before describing the background and the results obtained in this
paper, let us recall the denitions of SOP
n
hierarchy, starting with the
more classical concepts introduced in [Sh500].
Let T be a complete rst order theory, C - the monster model of T
(a
- saturated model for
big enough).
Denition 1.1. (1) Let n 3. We say ( x, y) (with lg(x) = lg(y))
exemplies the strong order property of order n (SOP
n
) in T
if it denes on C a directed graph with innite indiscernible
chains and no cycles of length n.
(2) We say ( x, y) (with lg(x) = lg(y)) exemplies the strict order
property in T if it denes on C a partial order with innite
indiscernible chains.
Fact 1.2. For a theory T, strict order property =SOP
n+1
=SOP
n
for all n 3.
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

MORE ON SOP1 AND SOP2 3
Proof. The rst implication is trivial, for the other one see [Sh500],
Claim (2.6).
We also remind the reader the following equivalent denition of
SOP
3
:
Fact 1.3. T has SOP
3
if and only if there is an indiscernible sequence
a
i
: i < ) and formulae ( x, y), ( x, y) such that
(a) ( x, y), ( x, y) is contradictory,
(b) for some sequence
b
j
: j < ) we have
i j =[= [
b
j
, a
i
] and i > j =[= [
b
j
, a
i
].
(c) for i < j, the set ( x, a
j
), ( x, a
i
) is contradictory.
Proof. Easy, or see [Sh500], Claim (2.20).
Remark 1.4. Note that if in the previous denition = , we get the
strict order property.
Now we recall the denitions of SOP
1
, SOP
2
and related properties:
Denition 1.5. (1) T has SOP
2
if there is a formula ( x, y) which
exemplies this property in C, and this means:
There are a
C for
>
2 such that
(a) For every

2, the set ( x, a
) : l < is consistent.
(b) If ,
>
2 are incomparable, ( x, a
), ( x, a
) is in-
consistent.
(2) T has SOP
1
if there is a formula ( x, y) which exemplies this
in C, which means:
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

There are a
C, for
>
2 such that:
(a) for

2 the set ( x, a
n
) : n < is consistent.
(b) if 0)
>
2, then ( x, a
), ( x, a
1
) is incon-
sistent.
(3) NSOP
2
and NSOP
1
are the negations of SOP
2
and SOP
1
respectively.
(4) T has SOP
1
property in C, and this means:
there are a
:
>
2) in C
T
such that
(a) ( x, a
n
)
(n)
: n < is consistent for every

2,
where we use the notation
l
=
if l = 1,
if l = 0
for l < 2.
(b) If 0)
>
2, then ( x, a
), ( x, a
) is inconsis-
tent.
(5) T has SOP
2
property in C, and this means:
there is n < and a sequence
a

: =
0
, . . .
n1
),
0
1
. . .
n1

>
2 and lg(
i
) successor
such that
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

(a) for each

2, the set
( x, a

) : = (
0
+ 1), (
1
+ 1), . . . (
n1
+ 1))
and
0
<
1
< . . .
n1
<
is consistent
(b) for every large enough m, if h is a 1-to-1 function from
n
m into
>
2 preserving and (incomparability)
then ( x, a

) : for some
n
m we have = h( ) :
n) is inconsistent.
Fact 1.6. (1) For a theory T, SOP
3
=SOP
2
=SOP
1
(2) T has SOP
1
if and only if it has SOP
1
Proof. See [DzSh692].
It is still not known whether the implications in 1.6(1) are strict, but
for now we investigate each one of these order properties on its own.
In the second section we expand our knowledge on SOP
1
. We present
a rank function measuring type-denable squares, i.e. pairs of types
of the form (p( x), q( y)) and show the rank is nite for every such a pair
if and only if T does not have SOP
1
(if and only if T does not have
SOP
1
). In fact, if one calls a tree of parameters a
:
>
2 showing
that ( x, y) exemplies SOP
1
in C (as in the denition of SOP
1
) a
SOP
1
tree, the rank measures exactly the maximal depth of a tree
like this that can be built in C . We also show a small application of
the rank.
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

It is easy to see (see [DzSh692]) that if ( x, y) exemplies SOP
1
in C then it also exemplies the tree property, so T has SOP
1
=T
is not simple. We show that the implication is proper, i.e. nd an
example of a theory T which is not simple, but is NSOP
1
. This theory
which we call T
feq
, was rst dened in [Sh457], and is used in [Sh500]
as an example of an NSOP
3
non-simple theory. Here we use a slightly
dierent denition of the same theory, as given in [DzSh692].
Denition 1.7. (1) T
feq
is the following theory in the language
Q, P, E, R, F
(a) Predicates P and Q are unary and disjoint, and (x) [P(x)
Q(x)],
(b) E is an equivalence relation on Q,
(c) R is a binary relation on QP such that
[xRz & y Rz & xE y] = x = y.
(so R picks for each z Q(at most one) representative of any E-equivalence
class).
(d) F is a (partial) binary function from Q P to Q, which
satises
F(x, z) Q & (F(x, z)) Rz & xE (F(x, z)) .
(so for x Q and z P, the function F picks the representative of the
E-equivalence class of x which is in the relation R with z).
(2) T
feq
is the model completion of T
feq
.
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

If the reader thinks about the denition above, they will nd out
that T
feq
is just the model completion of the theory of innitely many
(independent) parameterized equivalence relations. The reader can also
compare between the denition of T
feq
here and in [Sh457]. As we have
already mentioned, it was shown in [Sh500] that this theory does not
have SOP
3
(but is not simple). Here we prove an (a priori) stronger
result: T
feq
does not have SOP
1
.
In the third section we deal with
-maximality (see the beginning

of the section for denitions). For a theory T to be
-maximal means
to be complicated. In a way, it means that it is hard to make its models
-saturated.
The motivation for considering this property comes from Classica-
tion Theory and the search for dividing lines. The authors believe
that a good property of a theory T should have several characteri-
zations of dierent types, both internal (something happens in the
monster model of T, such as SOP
n
) and external (how T compares
to other theories, such as maximality in a certain order). Although
all the approximations to the strict order property (including SOP
n
and the strict order property itself) seem very natural syntactic inter-
nal denitions, no external property is known to characterize any of
them. There are natural conjectures, though. The following question
was partly guiding our current work:
Question 1.8. Does
-maximality characterize either SOP

3
or SOP
2
,
maybe both?
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

There are several indications that the answer should be positive.
It had been already known before our work that
-maximality lies
strictly above the tree property (non-simplicity): Dzamonja and
Shelah showed in [DzSh692] that T
feq
(which is not simple) is not
-maximal. The question where exactly above non-simplicity this

property lies is still open, but we narrow the possibilities down signi-
cantly. It follows from our results here that SOP
3
=
-maximality
= SOP
2
. We also obtain a local version of the reversed direction of
the second implication.
Our analysis also provides an alternative proof of the fact that T
feq
is not
-maximal, Theorem 1.17 in [DzSh692]: no NSOP

2
theory is
-maximal, and T
feq
is NSOP
1
, therefore NSOP
2
. So by bringing
the internal and the external dividing lines close together, we also
give many examples of non-simple theories which are not
-maximal,
T
feq
being a particular case. See also Discussion 3.18.
Let us now give more details concerning some results in the paper
and explain how exactly they t in the general picture. In [Sh500] it
was stated that SOP
3
implies
-maximality, but the proof there is

not complete: it is shown that every theory with SOP
3
is
-above
T
tr
, the model completion of the theory of trees. The rst theorem in
section 3, Theorem 3.5, lls the missing part, showing explicitly that
T
tr
is
-maximal for every >

0
. This also continues [Sh:c], chapter
VI, where Keislers order, a relative of
, is studied.
One of the reason for giving an explicit proof for Theorem 3.5 here
was to provide more tools for strengthening the result above to SOP
2
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

theories, i.e. showing that SOP
2
=
-maximality. A step in this

direction is Theorem 3.11 where we show a local version: if a for-
mula exemplies SOP
2
in T, then the pair (T, ) is
-above the
pair (T
tr
, y < x) for every regular > [T[ (again, see the beginning of
section 3 for precise denitions). This result, although interesting on
its own, is insucient for global
-maximality of SOP
2
theories,
as explained in Discussion 3.17. Nevertheless, combined with Theo-
rem 3.5 and its proof, it gives more information on the behavior of
SOP
2
theories and
-order altogether.
As for the other direction (
-maximality = SOP
2
), we provide a
complete proof, based on several related results achieved by Dzamonja
and the rst author, who showed in [DzSh692] that a property similar
to
-maximality (which also follows from
-maximality for some

under certain set theoretic conditions) implies SOP
2
. One of the ques-
tions left open in [DzSh692] is the connection between SOP
2
and the
SOP
n
hierarchy. Of course, it would be natural to connect between
SOP
2
and SOP
2
, and indeed we prove here that these two properties
are equivalent for a theory T (not necessarily for a formula), Theo-
rem 3.13.
So we can conclude SOP
3
=
-maximality =SOP
2
, while very
little is known at this point concerning implications in the other direc-
tions.
The following denitions and facts are going to be very useful.
In [DzSh692] two notions of tree indiscernibility were dened. We
recall the denitions:
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

Denition 1.9. (1) Given an ordinal and sequences
l
=
l
0
,
l
1
, . . . ,
l
n
l
)
for l = 0, 1 of members of
>
2, we say that
0

1

1
i
(a) n
0
= n
1
,
(b) the truth values of
l
k
3
l
k
1

l
k
2
,
l
k
1

l
k
2
l
k
3
, (
l
k
1

l
k
2
) 0)
l
k
3
,
for k
1
, k
2
, k
3
n
0
, do not depend on l.
(2) We say that the sequence a
:
>
2) of C (for an ordinal )
are 1-fully binary tree indiscernible (1-fbti) i whenever
0

1

1
are sequences of elements of
>
2, then
a

0
=: a
0
0
. . . a
0
n
0
and the similarly dened a

1
, realize the same type in C.
(3) We replace 1 by 2 in the above denition i (
l
k
1
l
k
2
) 0)
l
k
3
is omitted from clause (b) above.
We will need the following fact proved in [DzSh692], (2.11):
Fact 1.10. If t 1, 2 and
:
>
2) are given, and , then
we can nd a
:
>
2) such that
(a) a
:
>
2) is t-fbti,
(b) if =
m
: m < n), where each
m

>
2 is given, and is a
nite set of formulae of T, then we can nd
m

>
2 (m < n)
such that with =:
m
: m < n), we have
t
and the
sequences a

and

b

, realize the same -types.
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

Convention 1.11. We work with a complete rst order theory T,
let C be its monster model (saturated in some very big
). Let
L = L (T) (the language of T). Every formula we mention is an
L-formula, maybe with parameters from C.
2. More on SOP
1
SOP
1
was introduced by Dzamonja and Shelah as an intermediate
property between simplicity and SOP
3
. A natural question is: does the
class NSOP
1
coincide with either simple theories or NSOP
3
? Here we
give a negative answer to the rst question above. The answer for the
second one is still not known.
Theorem 2.1. T
feq
does not have SOP
1
.
Proof. Suppose there exists ( x, y) with g( x) = n, g( y) = m, and
a
:
>
2) in
m
C which exemplify SOP
1
in C (C is the monster
model of T
feq
). Without loss of generality, (by Fact 1.10) a
:
>
2)
if 1-full tree indiscernible. Also, by elimination of quantiers, we may
assume that ( x, y) is quantier free. As the only function symbol in
the language is F and F
C
has the property F
C
(F
C
(x, z), y) = F
C
(x, y)
for all z, we will also assume wlog that x and y in ( x, y) are closed
under F and ( x, y) gives the full diagram of x y. We shall regard
x as x
0
, . . . , x
n1
), y as y
0
, . . . , y
m1
), a
as a
0
, . . . , a
m1
).
By the denition of SOP
1
, there exist e = e
0
, . . . , e
n1
),

d =
d
0
, . . . , d
n1
) in
n
C s.t.
C [= ( e, a

) ( e, a
0
) ( e, a
00
)
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

and
C [= (
d, a

) (
d, a
1
)
Denote = 00). Let B = C a
a
1
. By our assumptions,
there exists a model N
0
whose universe is x a
, extending C a
,
whose basic diagram is ( x, a
). Similarly, there exists a model N

1
with
universe x a
1
) and basic diagram ( x, a
1
). We shall amalgamate
B, N
0
and N
1
into a model of T
feq
, N. This will immediately give a
contradiction: rst, extend N to N
[= T
feq
, then amalgamate N
and C over B into some C

+
[= T
feq
. By model completeness of T
feq
,
C C
+
, but C
+
[= x(( x, a
) ( x, a
1
)), which is a contradiction to
the denition of SOP
1
.
It is left, therefore, to show that we can dene on [N
0
[ [N
1
[ a
structure which will be a model of T
feq
, extending B.
We dene N as follows:
[N[ = [N
1
[ [N
2
[, P
N
= P
N
1
P
N
2
, Q
N
= Q
N
1
Q
N
2
.
Note that the diagram of x in N
0
is the same as the diagram of x
in N
1
(both implied by ( x, y), and the diagrams of a
, a
1
in N
i
are
the same as in C, hence the same as in B. Therefore, P
N
and Q
N
are
well dened and give a partition of [N[. Also, so far N extends B (as
a structure).
Considering E and R, we dene
R
N
= R
N
1
R
N
2
R
B
E
N
= E
N
1
E
N
2
E
B
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

Once we have proven the following claims, we will be able to dene F
N
in a natural way, and in fact will be done.
Claim 2.1.1. E
N
is an equivalence relation on Q
N
, extending E
B
.
Claim 2.1.2. R
N
is a two-place relation on N, R
N
P
N
Q
N
,
satisfying:
for every y P
N
and every equivalence class C of E
N
, there exists
a unique z C such that (y, z) R
N
.
Proof of 2.1.1. The only nonobvious thing is transitivity. We check
two main cases, all the rest are either similar or trivial.
(1) Assume x
i
E
N
a
j
, x
i
E
N
a
k
1
for some i, j, k. We want to show
a
j
E
N
a
k
1
. It is enough to see a
j
E
C
a
k
1
. We will write E instead
of E
C
.
N [= x
i
Ea
j
N
0
[= x
i
Ea
j
( x, y) x
i
Ey
j
. Similarly,
( x, y) x
i
Ey
k
, and we get (by the choice of e,

d
n
C)
e
i
Ea
j
, e
i
Ea
j

, e
i
Ea
k
, e
i
Ea
k

, d
i
Ea
j
1
, d
i
Ea
j

, d
i
Ea
k
1
, d
i
Ea
k

.
Now it is easy to see that all the above elements are E-equivalent
in C, in particular a
j
and a
k
1
, as required.
(2) Assume x
i
E
N
a
n
, a
k
1
E
N
a
n
, and we show x
i
E
N
a
k
1
, i.e. ( x, y)
x
i
Ey
k
. As (
d, a
1
) holds in C and ( x, y) gives a full diagram,
it will be enough to see d
i
Ea
k
1
.
We know that ( x, y) x
i
Ey
j
therefore e
i
Ea
j
, e
i
Ea
j

, d
i
Ea
j
1
, d
i
Ea
j

.
In particular, d
i
Ea
j
, but, by our assumption, a

j
Ea
k
1
, so we are done.
1
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

Proof of 2.1.2. Like in the previous lemma, the only nontrivial thing
to prove is the last part, and we will deal with two main cases.
(1) N [= (a
i
Ra
j
1
) (a
i
Rx
k
) (x
k
Ea
j
1
). We aim to show N [=
(x
k
= a
j
1
). We know:
()
1
C [= a
i
Ra
j
1
()
2
N
0
[= a
i
Rx
k
, therefore ( x, y) y
i
Rx
k
()
3
N
1
[= x
k
Ea
j
1
, therefore ( x, y) x
k
Ey
j
.
So we can conclude:
()
2
a
i

Re
k
, a
i

Rd
k
()
3
e
k
Ea
j

, d
k
Ea
j

e
k
Ed
k
.
As the above two relations hold in C, which is a model of
T
feq
, we get C [= e
k
= d
k
. Denote e
= e
k
= d
k
.
()
1
a
i
Ra
j
1
()
2
a
i
Re
()
1
e
Ea
j
1
Together (once again, C [= T
feq
) we get e
= a
j
1
, therefore
( x, y) x
k
= y
j
, so N
1
[= x
k
= a
j
1
, and we are done.
(2) N [= (x
i
Ra
j
1
) (x
i
Ra
k
) (a
k
Ea
j
1
) and we aim to show N [=
(a
k
= a
j
1
).
We know:
()
1
N
1
[= x
i
Ra
j
1
, so ( x, y) x
i
Ry
j
()
2
N
0
[= x
i
Ra
k
, so ( x, y) x
i
Ry
k
()
3
C [= a
k
Ea
j
1
Note that by indiscernibility of a
r
: r
w>
2) and ()
3
we get a
k
0
Ea
j
1
,
therefore a
k
0
Ea
k
. Now, by ()
2
, e
i
Ra
k
& e
i
Ra
k
0
. Therefore, by
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

C [= T
feq
, a
k
0
= a
k
. Now by indiscernibility
a
k
0
= a
k

, a
k
1
= a
k

So we get that all of the above are equal (and in fact a
k
r
1
= a
k
r
2
for all
r
1
, r
2

w>
2).
Now:
()
1
d
i
Ra
j
1
()
2
d
i
Ra
k
1
d
i
Ra
k
(as a
k
1
= a
k
)
()
3
a
k
Ea
j
1
.
By C [= T
feq
, we conclude a
k
= a
k
1
, which nishes the proof of the
lemma, and therefore the proof of the theorem.
2
Our next goal is to show that there is a rank function closely related
to being (N)SOP
1
. Let ( x, y) be a formula.
Denition 2.2. Given (partial) types p( x), q( y). By induction on n <
we dene when
rk
1
( x, y)
(p( x), q( y)) n :
If n = 0, this happens if both p( x), q( y) are consistent
For n + 1, the rank is n + 1 if for some c [=q( y), both
rk
1
( x, y)
(p( x) ( x, c), q( y)) n
and
rk
1
( x, y)
(p( x), q( y) ( x)(( x, y) ( x, c))) n.
We say rk
1
( x, y)
(p( x), q( y)) = i rk
1
( x, y)
(p( x), q( y)) n for all n.
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

We say the rank is 1 if it is not bigger or equal to 0.
Remark 2.3. (1) (Denability) Given formulae
1
,
2
and n < , the
statement rk
1
( x, y)
(
1
( x; a),
2
( x;
b)) n is a rst order formula with

parameters a,
b.
(2) (Finite Character) If rk
1
( x, y)
(p( x), q( y)) = n, then for some nite
p
0
( x) p( x) and q
0
( y) q( y) we have rk
1
( x, y)
(p
0
( x), q
0
( y)) = n.
(3) (Monotonicity) If p
and q
, then rk
1
( x, y)
(p
, q
) rk
1
( x, y)
(p
, q
).
(4) We can continue to dene when rk
1
( x, y)
(p( x), q( y)) for any
ordinal , but by the compactness theorem, part (1) (Denability) and
part (2) (the Finite Character) it follows that rk
1
( x, y)
(p, q) for
some i rk
1
( x, y)
(p, q) i rk
1
( x, y)
(p, q) = .
(5) If p
, and q
, then rk
1
(p
, q
) =rk
1
(p
, q
).
We aim to show that rk
1
(p( x), q( y))is nite for every p( x), q( y)(or,

equivalently, rk
1
( x = x, y = y) is nite) if and only if ( x, y) does not

exemplify SOP
1
in T. For this purpose we shall need another denition
and several easy claims.
Denition 2.4. Given (partial) types p( x)and q( y), we say that a
:

n
2 is a -SOP
1
tree for p( x)and q( y) (of depth n) if
(a) p( x)
(i)
( x, a
i
) : i < n is consistent for every
n
2.
(b) a
[= q( y) for all
n
2
(c) If , are in
n
2 satisfying 0) , then the set ( x, a
), ( x, a
)
is inconsistent.
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

Proposition 2.5. Suppose a
:
n
2 is a -SOP
1
tree for p( x)and
q( y) of depth n, and denote A
0
= a
: 0) , A
1
= a
: 1) .
Then
(1) A
1
is a -SOP
1
tree for p( x) ( x, a
) and q( y)
(2) A
0
is a -SOP
1
tree for p( x)and q( y) ( x)(( x, y) ( x, a
)).
Proof. The clauses (a) and (c) of the denition easily hold both for
A
1
and A
0
, so we should only check (b), which is also obvious for A
1
.
Therefore, were left to show that for every A
0
, a
[= ( x)(( x, y) ( x, a
)),
and this is clear by clause (c) of the denition ( a
:
n
2 is a -
SOP
1
tree, and ) 0 ).
Now we show the connection between the rank and SOP
1
trees.
Proposition 2.6. rk
1
(p( x), q( y)) n there exists a -SOP
1
tree
for p( x)and q( y)of depth n.
Proof. Both directions are proved by induction on n. The case n = 0 is
obvious. For n = m+1, the right-to-left direction follows immediately
by the induction hypothesis and 2.5. So we will elaborate more only
about the other direction, although it is also straightforward.
Suppose n = m+1 and rk
1
(p( x), q( y)) n. By the denition of the

rank and the induction hypothesis, for some c [=q( y), there are
(1) a -SOP
1
tree A
1
= a
1
:
m
2 for p( x) ( x, c) and
q( y)
(2) a -SOP
1
tree A
0
= a
0
:
m
2 for p( x)and q( y) ( x)(( x, y) ( x, c))
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

(both of depth m). We dene a tree a
:
n
2 by
a
= c
a
= a
for 0, 1
which is as required, i.e. a -SOP
1
tree for p( x)and q( y), since:
(a) of the denition obviously holds by (1) above.
(b) holds as c[=q( y).
(c) obviously holds by (2) above.
The following remark is obvious:

Remark 2.7. ( x, y) exemplies SOP
1
in T there exists a -SOP
1
tree for x = x and y = y of any depth.
So we can conclude the following
Theorem 2.8. A formula ( x, y) does not exemplify SOP
1
in T rk
1
( x = x, y = y) <
rk
1
(p( x), q( y)) < for every two (partial) types p( x)and q( y). More-
over, rk
1
( x = x, y = y) is exactly the maximal depth of a -SOP
1
tree
that can be built in C .
Corollary 2.9. T does not have SOP
1
T does not have SOP
1
rk
1
( x = x, y = y)
is nite for every formula ( x, y).
Now we show an application of the rank.
Theorem 2.10. Suppose that T satises NSOP
1
. Assume that
(a) M
1
M
2
C.
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

(b) p is a (not necessarily complete) type over M
2
, containing the
formula ( x,
) for some

b
M
2
M
1
.
Then for some nite q
tp(
/M
1
) at least one of the following holds:
(i) If

b M
1
realises q
( y) then ( x,
b) / p, or
(ii) If

b M
1
realises q
( y) then ( x,
b), ( x,
) is consistent.
In fact, all we need to assume for this Claim is that ( x, y) does not
exemplify SOP
1
in C.
Proof. Denote q = tp(b
/M
1
). As T is NSOP
1
,we have that rk
1
( x, y)
(p
M
1
, q) = n
< (certainly n
0). By the nite character of the rank,

we have that for some nite p
0
p M
1
and q
0
q,
rk
1
( x, y)
(p
0
, q
0
) = n
.
Hence for no c [= q
0
( y) do we have that both rk
1
( x, y)
(p
0
( x, c), q
0
)
n
and rk
1
( x, y)
(p
0
, q
0
( x)[( x, y) ( x, c)]) n
. In particular,
this holds for c =

b
(remember that

b
[= q and therefore certainly
[= q
0
). So
2.10.1.
If rk
1
( x, y)
(p
0
( x,
), q
0
) n
, then rk
1
( x, y)
(p
0
, q
0
( x)[( x, y)( x,
)]) < n
.
By Remark 2.3(1), there is a nite q
q such that
2.10.2.
b realises q
= rk
1
( x, y)
(p
0
( x,
b), q
0
) = rk
1
( x, y)
(p
0
( x,

b
), q
0
).
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

We aim to show that q
is as required.
Case 1. rk
1
( x, y)
(p
0
( x,

b
), q
0
) = n < n
.
We note that possibility (i) holds.
Namely, suppose

b realises q
,then rk
1
( x, y)
(p
0
( x,
b), q
0
) = n <
n
, so if ( x,
b) p, we obtain a contradiction with monotonicity of

the rank.
Case 2. rk
1
( x, y)
(p
0
( x,

b
), q
0
) = n
.
We shall show that (ii) holds.
Suppose otherwise, so let

b M
1
realise q
and ( x,
b), ( x,

b
) is
contradictory. By 2.10.2,
rk
1
( x, y)
(p
0
( x,
b), q
0
) = n
and by 2.10.1,
rk
1
( x, y)
(p
0
, q
0
( x)(( x, y) ( x,
b)) < n
.
We have that ( x)[( x, y) ( x,
b] q, hence q
0
( x)[( x, y)
( x,
b)] q, in contradiction with monotonicity and rk

1
( x, y)
(p M
1
, q) =
n
.
3. More on SOP
2
, SOP
3
and
-order
We try to nd a connection between the syntactic properties SOP
2
,SOP
3
and the semantic property of being
-maximal. Our guess is that
-maximality should be equivalent to one of the above order proper-

ties (maybe both), but all we prove here is SOP
3
=
-maximality
=SOP
2
. We also give a weaker local result in the other direction.
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

First we generalize the denitions from [DzSh692], of
-maximality,
making them local as well as global.
Denition 3.1. (1) For given (complete rst order theories) T
1
, T
2
and cardinals > , > [T
1
[ +[T
2
[ +
0
(a) T
1
<,<,,<
T
2
means that there is a (complete rst or-
der theory) T
and interpretations
1
,
2
of T
1
, T
2
in T
respectively, [T
[ < such that:

<,<,
T,
1
,
2
if M is a -saturated model of T
and M
=
M
[
]
for = 1, 2 and M
2
is -saturated (model of
T
2
), then M
1
is -saturated
(b) (T
1
,
1
( x, y))
<,<,<
(T
1
,
2
( x, y)) means that
( x, y)
L(
T
) and that there is a T
and interpretations
1
,
2
of
T
1
, T
2
in T
respectively, [T
[ < such that

<,<,
T
,
1
,
2
,
1
,
2
if
M is a -saturated model of T
and M
= M
[
]
for = 1, 2
and M
2
is (,
1
( x, y))-saturated (see 3 below), then M
i
is
(,
2
)-saturated.
(2) Instead of <
+
we may write , and instead of <
+
we
may write , instead of <
+
we may write . If we omit we
mean = , and if we write = 0 then -saturated becomes
the empty demand, if we omit we mean [T
1
[ +[T
2
[ +
0
and if
we omit and then we mean that = , = [T
1
[ +[T
2
[ +
0
.
(3) We say M is (, )-saturated when: if p ( x; a) : ( x; y)
, a
g( y)
M is nitely satisable of cardinality < then p is
realized in M. If = ( x, y) we may write ( x, y) instead
of .
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

(4) If T
1
, T
2
are not necessarily complete, then above T
is not
necessarily complete and we demand: if M
1
[= T
1
, M
2
[= T
2
then there is M [= T
such that M
[
]
[= Th(M
) for = 1, 2.
(5) We say T is
,
-maximal if [T
[ < T
,
T. We say
(T, ( x; y)) is
,
-maximal if [T
[ < &
( x
; y
)) L(
T
)
(T
( x
; y
))
,
(T, ( x; y)).
Denition 3.2. (1) T
tr
is the theory of trees (i.e. the vocabulary
is < and the axioms state that < is a partial order and y :
y < x is a linear order for every x), so T
tr
is not complete, and
let
tr
(x, y) = (y < x).
(2) T
tr
is the model completion of T
tr
.
(3) T
ord
is the theory of linear orders, T
ord
is its model completion
(i.e. the theory of dense linear order without endpoints).
We note connection to previous works and obvious properties
Proposition 3.3. (1) T
1
,,0
T
2
is T
1
,
T
2
of [DzSh692].
(2) T
1
,;<
T
2
implies T
1
,
T
2
of [Sh500].
(3)
,;,
has the obvious monotonicity properties: if T
1
<
1
,<
1
;<
1
,<
1
T
2
and
2

1
,
2

1
,
2

1
,
2

1
then T
1
<
2
,<
2
;<
2
,<
2
T
2
.
(4) T
,;,
T if [T[ < , > , .
(5) If is a limit cardinal, then T
1

<,<;<,<
T
2
i for every
1
< ,
1
we have
T
1
<,<
1
;<,<
T
2
.
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

(6) Similar results hold for (T
( x; y)).
Proof. Easy.
Proposition 3.4. (1) Assume T
1

<,<;<,<
T
2
. Then for any
theory T
, we can nd T
complete [T
[ < ([T
[
|(T
1
)|+|(T
2
)|
)
+
+
such that: for any interpretations
1
,
2
of T
1
, T
2
in T
re-
spectively the denition 3.1(1) of T
1
<,<;<,<
T
2
holds.
(2) Assume (T
1
), (T
2
) are disjoint. Then T
1
<,<;<,<
T
2
if for
any T T
1
T
2
there is T
T as demanded in Denition
3.1(1) for the trivial interpretations M
[
]
is the (T
)-reduct.
Proof. Easy, or see [DzSh692], Observation 1.4.
Now we will show that T
tr
is
-maximal for every big enough,

and conclude that SOP
3
=
-maximality. The last result appears

already in [Sh500], Theorem 2.9, but the proof is not complete - in fact,
the proof shows the following theorem:
Theorem 3.5. Any theory T, [T[ < , with SOP
3
is
-above T
ord
.
Proof. See [Sh500], (2.12).
Here we prove explicitly that T
tr
, and therefore T
ord
are maximal.
Theorem 3.6. T
tr
is
-maximal for any >

0
; the witness T
does
not depend on .
Proof. Let T be any complete theory, [T[ < and M
1
a model of T.
Let = (x, a) : (x, y) L(
T
), a
g( y)
(M
1
), so [[ = |M
1
|.
So M = (
>
, ) is a model of T
tr
and there is a model M
2
of T
tr
of
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

cardinality |M
1
| extending M such that every member of M
2
is below
some member of M.
Let be large enough such that M
1
, M
2
H() and we dene
B
expanding (H(), ) by P
1
= [M
1
[, P
2
= [M
2
[, P = [M[, Q
0
=
, <
1
=<
M
2
, <=<
1
P, m = a constant symbol for a set M
1
, R
B
= R
M
1
for R
T
(wlog (T) does not contain any other predicate mentioned
here)
Q = (
(x, a
) : < n) : M
1
[= x[
(x, a
)].
H is a partial unary function with domain Q and range P
1
, H(
(x, a
) :
< n)) satises
(x, a
) : < n, i.e. B
satises the formula

m [= (x)
<n

(x, a
).
Let T
= Th(B
), let
1
be the trivial interpretation of T in T
(the
restriction + reduct) and
2
= P
2
(x), x
0
<
1
x
1
) is an interpretation
of T
tr
. So T
,
1
,
2
does not depend on .
Now we assume B is a model of T
, N
1
= B
[
1
]
, N
2
= B
[
2
]
, N
3
=
(P
B
, <
B
) and we aim to show that (i) below implies (iii). We will rst
show that (i) (ii) and use this fact in the proof.
(i) N
2
is -saturated
(ii) in N
3
every branch has conality , equivalently: every in-
creasing sequence of length < has an upper bound
(iii) N
1
is -saturated.
Let us rst show (i) (ii). If a
i
: i < ) is <
N
3
-increasing, <
then it is <
N
2
-increasing hence has a <
N
2
-upper bound a but (x
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

P
2
)(y)(x <
1
y&P(y)) belongs to T
so there is b, a <
N
2
b P
N
= N
3
so b is as required.
So we can assume clause (i) and we shall prove (iii).
Before we proceed, let us note several easy but important properties
of B.
(a) We can talk inside B about a set being a model, (standard
coding of) a formula, a proof, etc. In particular, we can speak
about m (as a model) satisfying or not satisfying certain sen-
tences. Also, given a formula with free variables we can speak
about substitution of other variables or parameters into the for-
mula. Given s B which is a formula with free variables x,
we will allow ourselves to write s = s( x), and if B thinks that
substitution of a P
1
into s will turn it into a true sentence in
m as a model, we will write m [= s( a) or just s( a).
(b) B [= zQ
0
(z) z is a formula with one free variable with
parameters from P
1
. Moreover, suppose (x, a) is a formula
in L(
T
) s.t. a P
B
1
. B
and therefore B satisfy ( y

P
1
)(!s Q
0
) such that (x P
1
)(x, y) m [= s(x, y).
Let us denote by (x, a) this canonical encoding of (x, a)
in Q
B
0
.
(c) B [= sP(s) s is a nite sequence of members of Q
0
, i.e.
(n )(s : n Q
0
).
(d) For simplicity of notation, given s P
B
, we will write z s
instead of z Im(s
B
)
.
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

(e) For z P
B
, c P
B
1
, we write z(c) meaning (s z)s(c).
(f) For every (x, a) L(
T
) for a P
B
1
, there exists an element
of P
B
corresponding to the nite sequence (x, a)). We de-
note this element by (x, a)). Moreover, B [= x(P
1
(x)
(x, a)) Q((x, a))).
Subclaim 3.6.1. (1) Suppose B [= Q(z). Then B [= w(Q(w)
z < w) z(H(w)).
(2) Let (x, a) L(
T
) and suppose B [= xP
1
(x) (x, a). Then
B [= z(Q(z) (x, a)) < z) (H(z), a).
Proof. (1) Trivial as B
satises it.
(2) Let z
= (x, a)). First, Q(z
) holds by (f) above. By

(1), z
(H(z)) holds for each z Q

B
, z
< z. Now by (b)

and (f) above, B [= xP
1
(x) (z
(x) (x, a)). As

B [= Range(H) P
1
, we are done.
We now proceed with the proof (i) = (iii). So let p be a 1-type in

N
1
of cardinality < , so let p =
(x, a
) : < with < , a

N
1
. Without loss of generality p is closed under conjunction, i.e. for
every , < for some < we have
(x, a
) =
(x, a
(x, a
).
We shall now choose by induction on an element b
of N such
that
(A) b
P
B
= N
3
moreover b
Q
B
and < b
<
N
3
b
(B) if < then B [= (z)(Q(z) (b
z)
(H(z), a
))
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

(C) if < (but not necessarily < ) then B [= (z)[Q(z)(b

z) (y)(Q(y) z y
(H(y), a
))].
If we succeed then H
B
(b
) is as required.
Case 1: = 0.
Dene b
0
= ) (the element of P
B
corresponding to the empty
sequence). Clearly B [= Q(b
0
), i.e. the demand (A) holds. (B) holds
trivially. Why does (C) hold? Let < . B [= x
(x, a
) therefore
denoting z
(x, a
)), we have B [= Q(z
) b
0
< z
. Now we
nish by part (2) of the subclaim.
Case 2: = + 1.
B satises the sentence saying that for every Q and y P
1
there
exists an element of P that we denote by Conc
(, y) corresponding to
(x, y)). We dene b
= Conc
(b
, a
). Now we have to check

(A) - (C).
(A) By the induction hypothesis, clause (C) holds for b
and
(standing for b
and there). Therefore B [= z Q(b

z)
(H(z), a
)). But B
(and so B ) satises that y P

1
if
there exists z Q s.t.
(H(z), y) holds, then Conc
(z, y) is an
element of Q (as in B
the assumption means that there exists

an element of m satisfying all the formulae in z plus
(x, y)).
So we get the required.
(B) is clear as by the induction hypothesis,
(H(z), a
) holds for
every < , b
z (recall that b
). As for
(x, a
), B
clearly satises that for every z Q, y P

1
, if b = Conc
(z, y)
is in Q then
(H(z), y) holds z Q, b z.
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

(C) Let < . As p is closed under conjunctions, for some ,
(x, a
(x, a
) =
(x, a
). Now we apply clause (C) hold-

ing for b
to = and get z Q, b
z with H(z) satisfying

both
(x, a
) and
(x, a
). Once again using the satisfaction

by B of natural sentences, we show that b = Conc
(b
, a
) is in
Q, b
b and z Q which is above b,
(x, a
) holds, i.e. b
is as required.
Case 3: = limit.
By our present assumption, clause (i), and therefore clause (ii), hold.
Hence there is b P
B
which is an upper bound to b
: < . Now
B satises for every element z of P there is a y z which is in Q
and x z&Q(x) x y. Apply this to b for z and get b
for y. So
b
Q and < b
, as required in clauses (A) +(B) but not

necessarily (C).
Dene for each < a formula
(w, a
) = (z)(w z Q(z)
(y)(z y Q(y)
(H(y), a
)) Now we nd c
(for < ) such

that:
(a) c
Q
B
, c
b
(b)
(c
, a
) holds.
(c) under (a) + (b), the element c
is maximal.
Why do c
exist? B satises for every element s of P there is a

w s which satises
(w, a
), is in Q and (x s
(x, a
)Q(x))
(x w).
By the induction hypothesis we have:
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

< , < b
<
N
3
c
.
Clearly it suces to nd b
satisfying Q(b
) and b
<
N
3
b
<
N
3
c
for
< , < . As N
3
c : c b is linearly ordered, this follows from
N
2
being -saturated.
Proposition 3.7. (1) For every T
, there is T
, [T
[ =
[T
[ +
0
such that for every model B of T
we have
(a) for any , the following are equivalent
() if
1
is an interpretation of T
tr
in B (possibly with
parameters) then B
[
1
]
is
tr
-saturated
() if
2
ord
in B (possibly with
parameters) then B
[
2
]
is -saturated
(b) for any , the following are equivalent
() if
1
tr
in B (possibly with
parameters) then in B
[
1
]
, every branch with no last
element has conality
() if
2
ord
in B (possibly with
parameters) then in B
[
2
]
there is no Dedekind cut
(I
1
, I
2
) with both conalities < and at least one

0
.
Proof. Easy.
Corollary 3.8. (1) T
ord
is
-maximal.
(2) If [T[ < and T has SOP
3
then T is
-maximal.
Proof. (1) Follows from 3.7
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

(2) By (1) and 3.5.
Question 3.9. Is the other direction of 3.8 (2) true?

Remark 3.10. See Theorem 3.13 and Corollary 3.15 below for a proof of
a weaker version of the other direction: we get SOP
2
instead of SOP
3
.
We would like to know whether it is possible to weaken the assump-
tions of Corollary 3.8(2) to SOP
2
. The following theorem is a step in
this direction, showing a local version. See also Discussion 3.17.
Theorem 3.11. If T has SOP
2
as exemplied by ( x; y), then (T
tr
,
tr
(x; y))
(T, ( x; y)) for any [T[ +

0
regular.
Proof. We can nd a model M
1
of T
tr
and model M
2
of T and a
b

g( y)
M
2
for b M
1
such that:
() if M
1
[= b
0
< . . . < b
n1
then ( x, a
b
) : < n is satisable in
M
2
() if b
1
, b
2
are incomparable in M
1
then
M
2
[= ( x)(( x, a
b
1
)&( x, a
b
2
))
() for no

d
g( x)
(M
2
) is b M
1
: M
2
[= (
d, a
b
) unbounded
in M
1
(note that by () it is always linearly ordered in M
1
,
therefore () means that for each

d
g( x)
(M
2
), there exists an
element of M
1
which is above every b satisfying (
d, a
b
)).
[The construction of M
1
and M
2
is as follows: choose by
induction on n, (M
1,n
, M
2,n
, a
b
: b M
1,n
) : n < ) such that:
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

(a) M
1,n
is a model of T
tr
(b) M
2,n
is a model of T
(c) M
1,n
M
1,n+1
moreover, every branch of M
1,n
has an up-
per bound in M
1,n+1
(d) M
2,n
M
2,n+1
(e) a
b

g( y)
(M
2,n
) for b M
1,n
(f) clauses (), () hold
(g) if b M
1,n+1
and [b
M
1,n
M
1,n+1
[= (b < b
)] then
( x, a
b
) is not satised by any sequence from M
1,n
.
There is no problem to carry the denition.
Now M
1
=
n
M
1,n
, M
2
=
n
M
2,n
and a
b
: b M
1
) are as
required above.]
Now let be such that M
1
, M
2
H(), wlog
T
= (M
2
), < =
(T
tr
) = (M
1
) and are pairwise disjoint. Now we dene a model
B
0
.
Its universe is H() relation (membership)
P
1
= [M
1
[,
P
2
= [M
2
[
R = R
M
if R (M
), 1, 2 F
(for < g( y)) a partial unary

function such that: b M
1
F
(b) : < g( y)) = a

b
.
Let T
= Th(B
0
). For the obvious and

, T
is (T, T
tr
)-superior
and [T
[ = [T[ +
0
. Assume = cf() > [T
[.
So let B be a model of T
such that M
2
= B
[ ]
, the model of T
interpreted in it, is
+
-saturated. It will be enough to prove that
M
1
= B
[

]
satises: for every branch of conality there exists an
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

upper bound. So let b
i
: i < be <
M
1
-increasing let c
i
= F
B
(b
) :
< g( y)). Hence for any n < , i
0
< . . . < i
n1
< we have M
2
[=
( x)[
m<n
( x, c
i
)] because B
0
[= (z
0
, . . . , z
n1
)[
k<n
P
1
(z
k
) z
0
<
z
1
< . . . < z
n1
( x)
m<n
( x, F
(z
m
) : < g( y)))].
So ( x, c
i
) : i < is nitely satisable in M
2
hence some

d
g( x)
(M
2
) realizes it. Now we claim that b M
1
: B [= (
d, a
b
)
is bounded in M
1
: recall that by clause () B
0
satises: for every
x
g(y)
P
2
there exists z P
1
such that z is <
B
- above all the elements
w P
1
satisfying ( x, a
w
). Therefore B satises this sentence, and
applying it to

d
g( x)
(M
2
), we get b
1
- the required bound. As
for each i < , (
d, a
b
i
) holds, clearly B [= b
i
< b
for all i, and we are

done.
The next goal is to complete the proof started in [DzSh692] of the
fact that
-maximality implies SOP

2
. In [DzSh692] a property
-
maximality, which is closely related to
- maximality was dened,

and it was shown (Theorem 3.6 there) that every T which is
-
maximal for some (every) big enough regular , has an order property
similar to SOP
2
, that we call SOP
2
(see Denition 1.5). We will show
that SOP
2
is equivalent to SOP
2
(for a theory). This answers Question
(3.8)(3) from the original version of [DzSh692] (version 1 on the arXiv).
Discussion 3.12. In particular, Theorem 3.13 will lead to the following
conclusion: assuming that T is
+
- maximal for some regular
satisfying 2
=
+
, we get by [DzSh692], Claim (3.2) that T is
- maximal, so it has SOP
2
, and therefore SOP
2
. So we will obtain
-maximality implies SOP

2
, see Corollary 3.15.
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

Theorem 3.13. Let T be a theory.
(1) Suppose ( x, y) exemplies SOP
2
in T. Then ( x; y) exempli-
es SOP
2
in T as well.
(2) Suppose ( x, y) exemplies SOP
2
in T. Then for some k,
<k>
( x; y)
exemplies SOP
2
in T (where
<k>
( x; y
<k>
) =
<k
( x; y
)).
Proof. (1) is easy.
(2) Denote I
n
= : =
: n),
+1
; and

>
2. So
assume ( x; y) has SOP
2
as exemplied by n, a = a

: I
n
).
Without loss of generality a

: I
n
) is tree indiscernible
in the relevant sense: 0), 1) look the same over
(2 fbti from 1.9). We can assume this by 1.10 (for more
details, see [DzSh692], Claim (2.14)).
For

2 let p
= ( x, a

) : =
: < n),
<
+1
so
1
p
for

2 is consistent (in C
T
).
Let
=
(h, ) : h is a one-to-one mapping from

n
m to
>
2
preserving , and
n
m and there is
: ), h()

2 for
n
m
such that p
: is inconsistent
.
Now
2
is nonempty
[By the denition of SOP
2
, clause (b), choose =
n
m ]
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

Choose (h
) with [
[ of minimal cardinality and
) as there. By
1
clearly [
[ 2. So choose
0
,=
1
from
with
1
(= h(
0
) h(
1
)) being of maximal
length and let k
= g(
). We can nd
< suciently
large such that p
is inconsistent. We choose
by induction on i < for every

2, a sequence

>
2 by
= ),
<j>
=
).
Lastly for
>
2 , <> let
( x,
) be the conjunction
of
0
,
1
( x, a

) : =
: n),
+1
and ( n)[g(
) / [k
, g(
+ k
)]
(the last condition is empty if g() = 1)
.
In other words, we are taking the upper part of
that
looks like
0
or
1
after they split.
Now if

2 then
(x,
) :
is consistent as all its

members are conjunctions of formulas from
p
0
,
1
p
and this is consistent as otherwise (h
0
,
1
)
0
,
),
1
) belongs to for some
, thus contradicting
the choice of (h
), i.e. with minimal [
[.
Lastly if
0
,
1

>
2 are -incomparable then
( x;
0
),
( x;
1
)
is inconsistent: we know that
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

3.13.1.
0
,
1
( x, a

) : =
: n),
+1
( x, a

) : =
: n),
+1
.
is inconsistent (by the choice of (h
) and the choice

of
). Now, by the fact that
1
was chosen to be
maximal among other pairs in
, we see that if

0
=
0
: n), where for each ,

0

0
+1
and

1
=
1

1

1
+1
while

3
=
3

3
for some
0
,
then
3.13.2.

1

2

3

1

2

3
where
j
=
j
: n) and
=
j
, if lg(
j
) k
j
[lg(
j
) (
lg(
j
))], otherwise
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

In simpler words: we replace every
j
(an initial segment of
) whose length is bigger than k
(in particular, it is not

below any element in the image of
other than
j
itself ) by
an appropriate initial segment of
j
, and get a similar sequence
over the image of
0
,
1
.
Now, by indiscernibility of a
eta
), the denition of
( x,
), 3.13.1
and 3.13.2, we conclude
( x;
0
),
( x;
1
) is also inconsis-
tent.
Let us summarize the main results of this section.

Denition 3.14. (1) We call a theory T
-maximal if it is
-
maximal for every regular > [T[ +
0
.
(2) We call a pair of theory and formula (T, )
-maximal if it is
-maximal for every regular > [T[ +

0
.
Corollary 3.15. (1) If T has SOP
3
then it is
-maximal.
(2) If T is
-maximal then it has SOP

2
.
Proof. (1) Corollary 3.8.
(2) By [DzSh692] Claim 3.2, [DzSh692] Theorem 3.6 and Theo-
rem 3.13 above.
So we have shown SOP

3
=
-maximality = SOP
2
, and for
the second implication we also have a weak (local) converse, Theo-
rem 3.11. See Discussion 3.17 below.
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

Question 3.16. Is any of the two implications above reversible?
Discussion 3.17. Note that Theorem 3.11 is a step in (possibly) re-
versing the second implication above: we show that if T has SOP
2
exemplied by a formula , then the pair (T
tr
,
tr
) is
-below the pair

(T, ). By Theorem 3.6 (and quantier elimination), in order to obtain
SOP
2
=
-maximality it is enough to show that (T
tr
,
tr
,
tr
) is
-below (T, ) where is some fragment of the language of T. This

was our original motivation for proving Theorem 3.11, which is in a
sense a local or positive version of what we are interested in, but
right now it is unclear to us whether similar techniques will lead to the
desired global result.
One should remark that Theorem 3.11 is not weaker than the global
version since = , so it is really localized to a single formula, with
no use of negation (hence positive). Therefore, although it does not
quite do what we one would hope for, we nd Theorem 3.11 interesting
on its own.
Discussion 3.18. We would also like to point out that our analysis pro-
vides an alternative (in fact, in a sense a more conceptual) proof of
Theorem 1.17 and Conclusion 1.18 in [DzSh692]. Theorem 3.5 here
shows that T
tr
, and therefore T
ord
is maximal in
, and therefore is
-above T
feq
. By Theorem 2.1, T
feq
does not have SOP
1
(in partic-
ular, does not have SOP
2
), and so by Corollary 3.15, can not be
-
maximal. So T
feq
is strictly below T
tr
(and T
ord
) in
-ordering, which
is precisely the statement of Conclusion 1.18 in [DzSh692]. There is
no surprise here: in [DzSh692] it was shown that T
feq
is on the good
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

side of an external dividing line. Here we showed that it is also on
the good side of an internal syntactic dividing line, and brought
the internal and external lines close together. So our paper also
provides a generalization of section 1 of [DzSh692].
References
[DzSh692] M.Dzamonja and S.Shelah, On
-maximality, Annals of Pure and Ap-

plied Logic 125 (2004) 119-158.
[Sh457] S.Shelah, The Universality Spectrum: Consistency for more classes in
Combinatorics, Paul Erd os is Eighty, Vol. 1, 403-420, Bolyai Society
Mathematical Studies, 1993, Proceedings of the Meeting in honor of P.
Erd os, Keszthely, Hungary 7. 1993, an improved version available at
http://www.math.rutgers.edu/shelaharch
[Sh500] S.Shelah, Towards classifying unstable theories, Annals of Pure and Ap-
plied Logic 80 (1996) 229-255.
[Sh:c] S. Shelah, Classication theory and the number of nonisomorphic models,
2nd ed., North-Holland Publishing Co., Amsterdam, 1990.
Saharon Shelah, Mathematics Department, Hebrew University of
Jerusalem, 91904 Givat Ram, Israel,
Saharon Shelah, Department of Mathematics, Hill Center-Busch
Campus Rutgers, The State University of New Jersey 110 Frelinghuy-
sen Rd Piscataway, NJ 08854-8019, USA
Alexander Usvyatsov, UCLA Mathematics Department, Box 951555,
Los Angeles, CA 90095-1555
8
4
4

r
e
v
i
s
i
o
n
:
2
0
0
8
-
0
2
-
0
4

m
o
d
i
f
i
e
d
:
2
0
0
8
-
0
2
-
0
6

Alexander Usvyatsov, Universidade de Lisboa, Centro de Matem atica
e Aplicac oes Fundamentais, Av. Prof. Gama Pinto,2, 1649-003 Lisboa,
Portugal

Saharon Shelah and Alexander Usvyatsov - More On SOP1 and SOP2

Enviado por

Dados do documento

Descrição original:

Título original

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Saharon Shelah and Alexander Usvyatsov - More On SOP1 and SOP2

Enviado por

Direitos autorais:

Formatos disponíveis

8

among complete rst order

- saturated model for

-maximality (see the beginning

-maximality characterize either SOP

-maximal. The question where exactly above non-simplicity this

-maximal, Theorem 1.17 in [DzSh692]: no NSOP

-maximality, but the proof there is

-maximal for every >

-maximality. A step in this

-maximality (which also follows from

-maximality for some

). Similarly, there exists a model N

and C over B into some C

, but, by our assumption, a

b)) n is a rst order formula with

(p( x), q( y))is nite for every p( x), q( y)(or,

( x = x, y = y) is nite) if and only if ( x, y) does not

(p( x), q( y)) n there exists a -SOP

(p( x), q( y)) n. By the denition of the

The following remark is obvious:

( x = x, y = y) is exactly the maximal depth of a -SOP

0). By the nite character of the rank,

[= q and therefore certainly

b) p, we obtain a contradiction with monotonicity of

b)] q, in contradiction with monotonicity and rk

-maximal. Our guess is that

-maximality should be equivalent to one of the above order proper-

[ < such that:

) and that there is a T

[ < such that

-maximal for every big enough,

-maximality. The last result appears

-maximal for any >

satises the formula

and therefore B satisfy ( y

= (x, a)). First, Q(z

) holds by (f) above. By

(H(z)) holds for each z Q

< z. Now by (b)

(x) (x, a)). As

We now proceed with the proof (i) = (iii). So let p be a 1-type in

) : < with < , a

(B) if < then B [= (z)(Q(z) (b

)), we have B [= Q(z

(x, y)). We dene b

). Now we have to check

and there). Therefore B [= z Q(b

(and so B ) satises that y P

(H(z), y) holds, then Conc

the assumption means that there exists

clearly satises that for every z Q, y P

). Now we apply clause (C) hold-

z with H(z) satisfying

). Once again using the satisfaction

b and z Q which is above b,

, as required in clauses (A) +(B) but not

(for < ) such

exist? B satises for every element s of P there is a

Question 3.9. Is the other direction of 3.8 (2) true?

(T, ( x; y)) for any [T[ +

(for < g( y)) a partial unary