5

2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


Random Graphs in the Monadic Theory of Order.
BY
SHMUEL LIFSCHES and SAHARON SHELAH*
Institute of Mathematics, The Hebrew University, Jerusalem, Israel
ABSTRACT
We continue the works of Gurevich-Shelah and Lifsches-Shelah by showing that it is consistent
with ZFC that the rst-order theory of random graphs is not interpretable in the monadic theory
of all chains. It is provable from ZFC that the theory of random graphs is not interpretable in
the monadic second order theory of short chains (hence, in the monadic theory of the real line).
0. Introduction
We are interested in the monadic theory of order the collection of monadic sentences that are
satised by every chain (= linearly ordered set). The monadic second-order logic is the fragment of
the full second-order logic that allows quantication over elements and over monadic (unary) pred-
icates only. The monadic version of a rst-order language L can be described as the augmentation
of L by a list of quantiable set variables and by new atomic formulas t X where t is a rst order
term and X is a set variable.
It is known that the monadic theory of order and the monadic theory of the real line are at
least as complicated as second order logic ([GuSh2], [Sh1]). The question that we are dealing with
in this paper is related to the expressive power of this theory: what can be interpreted in it?
In our notion of (semantic) interpretation, interpreting a theory T in the monadic theory of
order is dening models of T in chains. Some problems about the interpretability power of the
monadic theory of order, which is a stronger criterion for complicatedness, have been raised and
answered. For example, second order logic was shown to be even interpretable in the monadic
theory of order ([GuSh3]) but this was done by using a weaker, non-standard form of interpretation:
into a Boolean valued model.
* The second author would like to thank the U.S.Israel Binational Science Foundation for par-
tially supporting this research. Publ. 527
1
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


Using standard interpretation ([GMS]) it was shown that it is consistent that the secondorder
theory of
2
is interpretable in the monadic theory of
2
(hence in the monadic theory of well
orders). On the other hand, by [GuSh], Peano arithmetic is not interpretable in the monadic theory
of short chains, (chains that do not embed (
1
, <) and (
1
, >)) and in particular in the monadic
theory of the real line. In [LiSh] we lled the gap left by the previous results and showed that it is
not provable from ZFC that Peano arithmetic is interpretable in the monadic theory of order.
Here we replace Peano arithmetic by a much simpler theory the theory of random graphs,
and obtain the same results by proving:
Theorem. There is a forcing notion P such that in V
P
, the theory of random graphs is not
interpretable in the monadic second-order theory of chains.
In fact we show that the model V
P
in which Peano arithmetic is not interpreted is a model in
which the theory of randomgraphs is not interpreted (an exact formulation of the non-interpretability
theorem is given in section 2).
The proof is similar in its structure to the proof in [LiSh]: we start by dening, following [Sh],
our basic objects of manipulation - partial theories. Next, we present the notion of interpretation
and the main theorem. We show in 3 that an interpretation in a chain C concentrates on an
initial segment D C called a major segment. One of the main dierences from [GuSh] and [LiSh]
is that the notion of a major segment is not as sharp as there; this results in the need to apply more
complicated combinatorial arguments.
The most widely used idea in the proof is applying the operation of shuing subsets X, Y C:
given a partition of C, S
j
: j J) and a subset a J, the shuing of X and Y with respect to J
and a is the set:

ja
(X S
j
)

ja
(Y S
j
). One of the main results in [LiSh] was to show that
this operation preserves partial theories; this is stated and used here as well.
To prove the main theorem we try to derive a contradiction from the existence of an interpre-
tation in a chain (C, <) V
P
. We start by making two special assumptions: that C itself is the
minimal major initial segment, and that C is an uncountable regular cardinal. The spirit of the
proof and main tools are similar to [LiSh], but some of the techniques have to be more tortuous.
The proof in this case contains all the main ingredients and disposing of the special assumptions is
essentially a formality.
Although we use many denitions and techniques from [Sh], [GuSh] and [LiSh] we have tried to
make this paper as self contained as possible. The only main proof we have omitted is that of the
theorem on preservation of partial theories under shuings, as its proof is quite long and involves
ideas that are not directly related to this paper.
1. Composition and Preservation of Partial Theories
In this section we dene formally the monadic theory of a chain and our main objects of interest:
its nite approximations (partial theories). We state the useful properties of partial theories, namely
the composition theorem and the theorem about preservation under shuing.
2
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


The monadic theory of a chain is dened to be the rst order theory of its power set.
Denition 1.1. Let (C, <) be a chain. The monadic second-order theory of C is the rst-
order theory of the model
C
mon
= (T(C) ; , <

, EM, SING)
where T(C) is the power set of C, < and are binary relations, SING and EM are unary relations
and:
(i) C
mon
[=SING(X) i X is a singleton,
(ii) C
mon
[= X <

Y i X = x, Y = y (where x, y C) and C [= x < y,

(iii) C
mon
[=EM(X) i X = ,
(iv) is interpreted as the usual inclusion relation between subsets of C.
Remark. We denote the rst order language above by L(mon). However we will be slightly
informal about that and identify it with the monadic version of the rst-order language of or-
der, L. Now each L can be translated to a rst-order formula

L(mon) by the rules:

(x)(x) (individual quantication) will be translated to (X)[SING(X) &

(X)] and x Y to
SING(X) &(X Y ). So when we write C [= (for L) we mean C
mon
[=

and x < y is
translated as X <

Y .
Notations 1.2. We denote individual variables by x, y, z and set variables by X, Y, Z. a, b, c
are elements and A, B, C are sets. a and

A denote nite sequences having lengths lg( a) and lg(

A).
We will write a C and

A C instead of a
lg( a)
C or

A
lg(

A)
T(C), we may also write a
0
a
or A
0


A.
Next is the denition of the partial n-theory of

A in C
Denition 1.3. Let (C, <) be a chain and

A C. We dene
t = Th
n
(C;

A)
by induction on n:
for n = 0: t = (

X) : L(mon), quantier free, C
mon
[= (

A)
for n = m+ 1: t = Th
m
(C;

A

B) : B C.
Lemma 1.4. (A) For every formula (

X) L there is an n such that from Th
n
(C;

A) we
can decide eectively whether C [= (

A). We call the minimal such n the depth of and write
dp() = n.
(B) For every n and l there is a nite set of monadic formulas (eectively computable from n and
l) (n, l) =
m
(

X) : m < m

, lg(

X) = l L such that for any chains C, D and

A C,

B D
of length l the following hold:
(1) dp(
m
(

X)) n for m < m

,
(2) Th
n
(C;

A) can be computed from m < m

: C [=
m
[

A],
(3) Th
n
(C;

A) = Th
n
(D;

B) i for every m < m

, C [=
m
[

A] D [=
m
[

B].
3
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


Proof. In [Sh], Lemma 2.1.

Denition 1.5. When (n, l) is as in 1.4(B), for each chain C and

A C of length l we can
identify Th
n
(C;

A) with a subset of (n, l). Denote by T
n,l
the collection of subsets of (n, l) that
arise from some Th
n
(C;

A) and call it the set of formally possible (n, l)-theories.
Remark. For given n, l N, each Th
n
(C;

A) is hereditarily nite, (where lg(

A) = l, C is a
chain), and we can eectively compute the set of formally possible theories T
n,l
. (See [Sh], Lemma
2.2).
Denition 1.6. If (C, <
C
) and (D, <
D
) are chains then (C + D, <) is the chain that is
obtained by adding a copy of D after C (where < is naturally dened).
If (I, <) is a chain and (C
i
, <
i
) : i I) is a sequence of chains then

iI
(C
i
, <
i
) is the chain
that is the concatenation of the C
i
s along I equipped with the obvious order.
Given

A = A
0
, . . . , A
l1
) and

B = B
0
, . . . , B
l1
) we denote by

A

B the sequence A
0

B
0
, . . . , A
l1
B
l1
). The heavily used composition theorem for chains states that the partial
theory of a chain is determined by the partial theories of its convex parts.
Theorem 1.7 (Composition theorem for chains).
(1) If C, C

, D and D

are chains,

A C,

A

,

B D and

B

are of the same length

and if
Th
m
(C;

A) = Th
m
(C

;

A

)
and
Th
m
(D;

B) = Th
m
(D

;

B

)
then
Th
m
(C +D;

A

B) = Th
m
(C

+D

;

A

).
(2) If I is a chain and Th
m
(C
i
;

A
i
) = Th
m
(D
i
;

B
i
) for each i I (with all sequences of subsets
having the same length) then
Th
m
_

iI
C
i
;
i

A
i
_
= Th
m
_

iI
D
i
;
i

B
i
_
.
Proof. By [Sh] Theorem 2.4 (where a more general theorem is proved), or directly by induction
on m. See also theorem 1.9 below.

Using the composition theorem we can dene a formal operation of addition of partial theories.
4
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


Notation 1.8.
(1) When t
1
, t
2
, t
3
T
m,l
for some m, l N, then t
1
+ t
2
= t
3
means: there are chains C and D,
and

A C,

B D such that
t
1
= Th
m
(C; A
0
, . . . , A
l1
) & t
2
= Th
m
(D; B
0
, . . . , B
l1
) & t
3
= Th
m
(C +D;

A

B).
(By the composition theorem, the choice of C and D is immaterial).
(2)

iI
Th
m
(C
i
;

A
i
) is Th
m
_
iI
C
i
;
iI

A
i
), (assuming lg(

A
i
) = lg(

A
j
) for i, j I).
(3) If D is a sub-chain of C and

A C then Th
m
(D; A
0
D, A
1
D, . . .)) is abbreviated by
Th
m
(D;

A).
(4) For a < b C and

P C we denote by Th
n
(C;

P)
[a,b)
the theory Th
n
([a, b);

P [a, b)).
We conclude this part by giving a monadic version of the Feferman-Vaught theorem. Note that
the composition theorem is a consequence.
Theorem 1.9. For every n, l < there is m = m(n, l) < , eectively computable from n
and l, such that if
(i) I is a chain,
(ii) C
i
: i I) is a sequence of chains,
(iii) for i I,

Q
i
C
i
is of length l,
(iv) for t T
n,l
, P
t
:= i I : Th
n
(C
i
;

Q
i
) = t,
(v)

P := P
t
: t T
n,l
),
then Th
n
(

iI
C
i
;

Q
i
) is computable from Th
m
(I;

P).
Proof. This is theorem 2.4. in [Sh].

Next we dene semiclubs and shuings and we quote the important preservation theorem.
Denition 1.10. Let >
0
be a regular cardinal
1) We say that a is a semiclub subset of if for every < with cf() >
0
:
if a then there is a club subset of , C

such that C

a and
if , a then there is a club subset of , C

such that C

a = .
(Note that and are semiclubs and that a club J is a semiclub provided that the rst and
the successor points of J are of conality
0
. Also, if a is a semiclub then a is one as
well.)
2) Let X, Y , J =
i
: i < a club subset of , and let a be a semiclub of . We will
dene the shuing of X and Y with respect to a and J, denoted by [X, Y ]
J
a
, as:
[X, Y ]
J
a
=
_
ia
_
X [
i
,
i+1
)
_

_
ia
_
Y [
i
,
i+1
)
_
3) When

X,

Y are of the same length, we dene [

X,

Y ]
J
a
naturally.
4) We can naturally dene shuings of subsets of an ordinal with respect to a club J and a
semiclub a otp(J).
6) a-Th
n
(;

P) is Th
n
(;

P, a) where a is a semiclub.
5
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


The next theorem, which will play a crucial role in contradicting the existence of interpretations,
states that the result of the shuing of subsets of the same type is an element with the same partial
theory. The proof of the preservation theorem in 4 of [LiSh] requires some amount of computations
and uses some auxiliary denitions that are not material in the other parts of the paper. For example,
the partial theories WTh
n
(C;

P), ATh
n
(, (C;

P)) and a-WA
n
(C;

P) are used in the proof and even
the formulation of the theorem but we can avoid dening them by noticing that (for a large enough
m) a-Th
m
(C;

P) computes all these partial theories. We also avoid the denition of an n-suitable
club (which relies on ATh
n
). All the details can be found of course in [LiSh].
Theorem 1.11 (preservation theorem). Let

P
0
,

P
1
be of length l, n < and a
be a semiclub.
Then there are an m = m(n, l) < and a club J = J(n,

P
0
,

P
1
) such that if

X := [

P
0
,

P
1
]
J
a
then
()
_
a-Th
m
(;

P
0
) = a-Th
m
(,

P
1
)
_

_
Th
n
(;

P
0
) = Th
n
(;

P
1
) = Th
n
(;

X)
_
.
Moreover, there is t

= t

(

P
0
,

P
1
) T
n,l
such that, for every J with cf() =
0
,
()
_
a-Th
m
(;

P
0
) = a-Th
m
(;

P
1
)
_

_
Th
n
(;

P
0
)
[0,)
= Th
n
(;

P
1
)
[0,)
= Th
n
(;

X)
[0,)
= t

_
.
Proof. By [LiSh] 4.5, 4.12.

Denition 1.12. Let

P
0
,

P
1
be as above. Call a club J an n-suitable club for

P
0
and

P
1
if for every semiclub a , () and () of 1.11 hold.
Fact 1.13. For every nite sequence P=

P
i
:

P
i
, lg(

P
i
) = l, i < k) and for every n <
there is a club J that is n-suitable for every pair from P.
Proof. By [LiSh] 4.3, 4.4.

2. Random Graphs and Uniform Interpretations

The notion of semantic interpretation of a theory T in a theory T

is not uniform. Usually

it means that models of T are dened inside models of T

but the denitions vary with context.

In [LiSh] we gave the general denition of the notion of interpretation of one rst order theory in
another. In our case, in which we deal with interpreting a class of theories, another notion emerges,
that of a uniform interpretation.
First we dene the theory of K-random graphs:
6
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


Denition 2.1. Let 1 < K . An undirected graph ( = (G, R) is a K-random graph if
_
A
0
, A
1
G & [A
0
[, [A
1
[ < K & A
0
A
1
=
_

_
(x G)(a A
0
)(b A
1
)[xRa & xRb]
_
.
(When this holds we will say that x separates A
0
from A
1
).
Denition 2.2. (1) RG
K
is the theory of all K-random graphs (that is all the sentences, in
the rst-order language of graphs, that are satised by every K-random graph). RG
i
K
is theory of
all the innite graphs that are K-random.
(2)
K
is the class of all the K-random graphs, (clearly K < L
L

K
).
i
K
is the classes
of innite K-random graphs.
(3)
n
is the class
K

1,<K<
,
i
n
is
i
K

1,<K<
.
The next denition is the one used in [LiSh]. It is applicable in dealing with RG

, but will have

to be modied for dealing with nitely-random graphs.
Denition 2.3. An interpretation of a model ( of RG
K
in the monadic theory of a chain C
is a sequence of formulas in the language L of the monadic theory of order
1 = U(

X,

W), E(

X,

Y ,

W), R(

X,

Y ,

W))
where:
1) U(

X,

W) is the universe formula that says which sequences of subsets of C represent elements
of (. We denote by C
U
the set

X C : C [= U(

X,

W).
2) E(

X,

Y ,

W) is the equality formula, an equivalence relation on C
U
. We write

A

B when
C [= E(

A,

B,

W).
3) R(

X,

Y ,

W) is the interpretation of the graph relation, a binary relation on C
U
which respects
i.e. C [= R(

A,

B,

W) depends only on the E-equivalence classes of

A and

B.
4)

W C is a nite set of parameters allowed in the interpreting formulas.
5) C
U
/ , R)

= (.
Denition 2.4. Let 1 be an interpretation of ( in the monadic theory of a chain C.
The dimension of the interpretation, denoted by d(1), is lg(

X). We will usually assume without
loss of generality that lg(

W) = d(1) as well.
The depth of the interpretation, denoted by n(1), is maxdp(U), dp(E), dp(R).
Denition 2.5. Let RG

be one of the theories dened in 2.2(1) and

be the respective
class. We say that the monadic theory of order interprets RG

(or

) if there is a chain C, a random

graph (

and an interpretation 1 = U(

X,

W), E(

X,

Y ,

W), R(

X,

Y ,

W)) with C
U
/ , R)

= (.
7
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


Common notions of an interpretation of a theory T
1
in a theory T
2
demand that every model of
T
1
is interpretable in a model of T
2
(as in [BaSh]) or that inside every model of T
2
there is a denable
model of T
1
(see [TMR]). Here we seem to require the minimum: a single model is interpreted in a
single chain. This is often useful, but not always:
Fact 2.6. For every 1 < K < there is a chain C and a sequence 1 such that 1 is an
interpretation of a model of RG
i
K
(hence of RG
K
) in C. (That is, RG
K
, RG
i
K
, RG
n
, RG
i
n
are
interpretable in the monadic theory of order).
Proof. We shall demonstrate the construction for K = 2; the other cases are similar.
Let C = (, <), we will show that there is a one-dimensional interpretation of an innite model
of RG
2
in C without parameters. For that we have to dene U(X), E(X, Y ) and R(X, Y ). Let:
U(X) := [X = a & a > 1] [X = x, a, b & x 0, 1 & a, b > 1] ;
E(X, Y ) := U(X) & U(Y ) & X = Y ;
R(X, Y ) := U(X) & U(Y ) and either:
[X = a & Y = 0, a, b & a < b] or
[Y = a & X = 0, a, b & a < b] or
[X = b & Y = 1, a, b & a > b] or
[Y = b & X = 1, a, b & a > b] or
[X = a & Y = x, c, d & x 0, 1 & a , c, d] or
[Y = a & X = x, c, d & x 0, 1 & a , c, d].
Clearly everything is expressible in L and R(X, Y ) denes on X : (, <) [= U(X) a graph
relation that is 2-random.

Motivated by the previous fact we will dene now the suitable modication of the previous
denitions. The idea is to interpret, in a uniform way, an innite set of random graphs.
Denition 2.7. A uniform interpretation of
n
in the monadic theory of order is a sequence
C
K
, I,

W
K
) : K A
where
1) C
K
is a chain,
2) A is an innite subset of ,
3)

W
K
C
K
for K A,
4) I = U(

X,

Z), E(

X,

Y ,

Z), R(

X,

Y ,

Z)) is a sequence of formulas in L,
5) 1
K
:= U(

X,

W
K
), E(

X,

Y ,

W
K
), R(

X,

Y ,

W
K
)) is an interpretation of a model of RG
K
in C
K
for K A.
Given d and n in N there is only a nite number of possible interpretations 1 having dimension
d and depth n. The following is therefore clear:
8
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


Proposition 2.8. The following are equivalent:
(A) There is no uniform interpretation of
n
in the monadic theory of order.
(B) For every n, d N there is K

= K

(n, d) N such that if K K

and 1 is an interpretation
of some ( [= RG
K
in a chain C, then either d(1) > d or n(1) > n.
(C) For every sequence 1 = U(

X,

Z), E(

X,

Y ,

Z), R(

X,

Y ,

Z)) there is K

= K

(1) < such that

there are no chain C,

W C, K K

and (
K
such that U(

X,

W), E(

X,

Y ,

W), R(

X,

Y ,

W)
is an interpretation of ( in C.

Our main theorem has therefore the following form:

Theorem 2.9 (Non-Interpretability Theorem). There is a forcing notion P such that in
V
P
the following hold:
(1) RG

is not interpretable in the monadic theory of order.

(2) For every sequence of formulas 1 = U(

X,

Z), E(

X,

Y ,

Z), R(

X,

Y ,

Z)) there is K

< ,
(eectively computable from 1), such that for no chain C,

W C, and K K

does U(

X,

W),
E(

X,

Y ,

W), R(

X,

Y ,

W)) interpret RG
K
in C.
(3) The above propositions are provable in ZFC. if we restrict ourselves to the class of short chains.
Remark. As an -random graph is K-random for every K < , an interpretation of RG

is a uniform interpretation of
n
. Therefore clause (1) in the non-interpretability theorem follows
from clause (2).
3. Major and Minor Segments
From now on we will assume that there exists (in the generic model V
P
that is dened later) a
uniform interpretation 1 of
n
in the monadic theory of order. For reaching a contradiction we have
to nd a large enough K = K(1) < (a function of the depth and dimension of 1) and show that
no chain interprets a K-random graph by 1. The aim of this (and the next) section is to gather facts
that will enable us to compute an appropriate K. The main observation is that an interpretation in
a chain C concentrates on a segment (called a major segment). One of the factors in determining
the size of K will be the relation between the major segment and the other, minor, segment.
Context 3.1. 1 = U(

X,

W), E(

X,

Y ,

W), R(

X,

Y ,

W)) is an interpretation of a K-random
graph ( = (G, R) on a chain C.

W C are the parameters, d = d(1) = lg(

X) = lg(

W) is the
dimension of 1 and n = n(1) is its depth.
Denition 3.2. A G is big for (K
1
, K
2
) if there is B G with [B[ K
1
such that :
() for every disjoint pair A
1
, A
2
G B with [A
1
A
2
[ K
2
there is some x A (A
1
A
2
) that
separates A
1
from A
2
i.e.
__
yA1
xRy
_

__
yA2
xRy
_
.
9
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


When () holds we say that B witnesses the (K
1
, K
2
)-bigness of A.
Non-bigness is an additive property:
Proposition 3.3. Let A G be big for (K
1
, K
2
) and suppose that A =

i<m
A
i
. Then there
is an i < m such that A
i
is big for (K
1
+K
2
, K
2
/m).
Proof. Let B G ([B[ K
1
) witness the bigness of A. For i < m we will try to dene by
induction counter-examples for bigness, that is a set B
i
G and a function h
i
so that:
(1) [B
i
[ K
2
/m,
(2) B
i
G (B

j<i
B
j
),
(3) h
i
: B
i
t, f,
(4) for no x A
i
B
i
we have (y B
i
)[xRy h(y) = t].
Suppose we succeed. Let C
1
:=
_
x :
_
i
(x B
i
& h
i
(x) = t)
_
and C
2
:=
_
x :
_
i
(x B
i
& h
i
(x) =
f)
_
. But [C
1
C
2
[ K
2
, C
1
C
2
G B and of course C
1
C
2
= so by the assumption on
A there is some x A that separates C
1
from C
2
. Such an x belongs to some A
i
and it separates
C
1
B
i
from C
2
B
i
. This contradicts clause (4).
Therefore at some stage i < m we cant dene B
i
and look at B

:= B

j<i
B
j
. Now
[B

[ K
1
+ i K
2
/m K
1
+ K
2
and being unable to continue means: if B
i
G B

and
[B
i
[ K
2
/m then for every partition of B
i
to B
1
i
and B
2
i
there is some x A
i
B
i
such that
__
yB
1
i
xRy
_

__
yB
2
i
xRy
_
. In other words, A
i
is big for (K
1
+ K
2
, K
2
/m) (witnessed by B

)
as required.

Notation 3.4.

A C is called a representative if it represents an element of ( i.e. if
C [= U(

A,

W) (of course lg(

A) = d). The representatives

A,

B C are called equivalent and we
write

A

B if they represent the same element in ( i.e. if C [= E(

A,

B,

W). We use upper case
letters such as

X,

A,

U
i
to denote representatives. The corresponding lower case letters (x, a, u
i
) will
denote the elements of ( that are represented by the former. So e.g.

A

B a = b.
Denition 3.5.
1) A sub-chain D C is a segment if it is convex (i.e. x < y < z & x,z D y D).
2) A Dedekind cut of C is a pair (L, R) where L is an initial segment of C, R is a nal segment of
C, L R = and L R = C.
3) Let

A,

B C. We will say that

A,

B coincide on (resp. outside) a segment D C, if

AD =

BD
(resp.

A (C D) =

B (C D) ).
4) The bouquet size of a segment D C denoted by #(D) is the supremum of cardinals [S[ where
S ranges over collections of nonequivalent representatives coinciding outside D. Thus #(D) n i
there are nonequivalent representatives A
1
, A
2
, . . . , A
n
coinciding outside D.
Denition 3.5. Let D C be a segment
1) D is i

-fat if #(D) i

10
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


2) D is (K
1
, K
2
)-major if there is a set

U
i
: i < i

of representatives coinciding outside D, and

representing a subset of ( that is big for (K
1
, K
2
).
3) D is called (K
1
, K
2
)-minor if it not (K
1
, K
2
)-major.
We denote by M
1
the number [T
n,3d
[ (i.e. the number of possibilities for Th
n
(C;

X,

Y ,

Z)).
Proposition 3.7. Let (L, R) be a Dedekind cut of C. If L [R] is (K
1
, K
2
)-major then R [L]
is not K
3
-fat where K
3
= M
1
(K
1
+K
2
) + 1.
Proof. Suppose

A
i
: i < i
L
) demonstrate that L is (K
1
, K
2
)-major, i.e. they represent a
(K
1
, K
2
)-big set a
i
: i < i
L
) in ( and

A
i

R
=

A

. Assume towards a contradiction that

B
i
: i < K
3
)
demonstrate that R is K
3
-fat (i.e. i < j < K
3
b
i
,= b
j
and

B
i

L
=

B

). Dene an equivalence
relation E
L
on 0, 1, . . . , i
L
1 by:
iE
L
j Th
n
(L;

A
i
,

B

,

W) = Th
n
(L;

A
j
,

B

,

W).
By the denition of M
1
, E
L
has at most M
1
equivalence classes. By proposition 3.3 there is a
L

0, 1, . . . , i
L
1, an E
L
equivalence class, such that

A
i
: i a
L
represents a (K
1
+K
2
, K
2
/M
1
)-
big subset of (. Let B ( witness the (K
1
+ K
2
, K
2
/M
1
)-bigness of a
i
: i a
L
. Since
[B[ (K
1
+ K
2
) and K
3
= M
1
(K
1
+ K
2
+ 1) we can choose some j
1
, j
2
< K
3
with b
j1
, b
j2
, B
and with Th
n
(R;

A

,

B
j1
,

W) = Th
n
(R;

A

,

B
j2
,

W). Now by the composition theorem 1.7, and the
choice of a
L
and j
1
, j
2
we have for every i a
L
:
Th
n
(C;

A
i
,

B
j1
,

W) = Th
n
(L;

A
i
,

B
j1
,

W) + Th
n
(R;

A
i
,

B
j1
,

W) =
Th
n
(L;

A
i
,

B

,

W)+Th
n
(R;

A

,

B
j1
,

W) = Th
n
(L;

A
i
,

B

,

W)+Th
n
(R;

A

,

B
j2
,

W).
Therefore for every i a
L
C [= R(

A
i
,

B
j1
,

W) R(

A
i
,

B
j2
,

W).
Since b
j1
, b
j2
, B we get a contradiction to A is (K
1
+K
2
, K
2
/M
1
)-big as witnessed by B.

Notation 3.8. Let M

2
be [T
n,2d
[, M
3
be M
1
+ 1 (= [T
n,3d
[ + 1) and M
4
be such that for
every colouring f: [M
4
]
3
0, 1, . . . , 6 there is a homogeneous subset of 0, 1, . . . , M
4
1 of size
M
3
, where [M
4
]
3
is i, j, k) : i < j < k < M
4
. (M
4
exists by Ramsey theorem).
The main lemma states that in every Dedekind cut one segment is major. Now we have to make
an assumption on the degree of randomness of (.
Lemma 3.9. Assume K > (M
3
)
2
(K is from K-random). Let (L, R) be a Dedekind cut of
C. Then either L or R is (K
1
, K
2
)-major where K
1
= K +
K
(M2)
2
and K
2
=
K
(M2)
2
M4
.
Proof. ( is (0, K)-big and let

U
i
: i < i

be a list of representatives for the elements of (.

Dene a pair of equivalence relations E
0
L
and E
0
R
on i

= [([ by:
_
iE
0
L
j Th
n
(L;

U
i
,

W) = Th
n
(L;

U
j
,

W)
_ _
iE
0
R
j Th
n
(R;

U
i
,

W) = Th
n
(R;

U
j
,

W)
_
.
11
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


By the denition of M
2
each relation has M
2
equivalence classes; therefore by 3.3 there is a subset
A
1
i

and pair of theories (t

1
, t
2
) such that u
i
: i A
1
is (K,
K
(M2)
2
)-big and
i A
1

_
Th
n
(L;

U
i
,

W) = t
1
& Th
n
(R;

U
i
,

W) = t
2

.
Denote by

X

Y the tuple (

X
L
) (

Y
R
).
() For i, j A
1
we have C [= U(

U
i

U
j
,

W) (hence

U
i

U
j
is a representative).
Why? Because C [= U(

U
i
,

W) and by the composition theorem
Th
n
(C;

U
i
,

W) = Th
n
(L;

U
i
,

W) + Th
n
(R;

U
i
,

W) = t
1
+t
2
=
Th
n
(L;

U
i
,

W) + Th
n
(R;

U
j
,

W) = Th
n
(C;

U
i

U
j
,

W).
Dene a pair of relations E
L
and E
R
on

U
i
: i A
1
by:

U
i
E
L

U
j
(r A
1
)
_

U
i

U
r


U
j

U
r

U
i
E
R

U
j
(l A
1
)
_

U
l

U
i


U
l

U
j

()

U
i
E
L

U
j
(r A
1
)
_

U
i

U
r


U
j


U
r
_
&

U
i
E
R

U
j
(l A
1
)
_

U
l


U
i


U
l


U
j
_
.
Why? Suppose

U
i
E
L

U
j
,

U
i


U
r


U
j


U
r
and let r
1
A
1
.
Now Th
n
(R;

U
r
,

W) = t
2
= Th
n
(R;

U
r1
,

W) hence Th
n
(R;

U
r
,

U
r
,

W) = Th
n
(R;

U
r1
,

U
r1
,

W). By
the composition theorem
Th
n
(C;

U
i


U
r1
,

U
j


U
r1
,

W) =
Th
n
(L;

U
i
,

U
j
,

W)+Th
n
(R;

U
r1
,

U
r1
,

W) = Th
n
(L;

U
i
,

U
j
,

W)+Th
n
(R;

U
r
,

U
r
,

W) =
Th
n
(C;

U
i


U
r
,

U
j


U
r
,

W).
Therefore Th
n
(C;

U
i

U
r1
,

U
j

U
r1
,

W) = Th
n
(C;

U
i

U
r
,

U
j

U
r
,

W) and hence

U
i

U
r


U
j


U
r


U
i


U
r1


U
j


U
r1
.
() [A
1
/E
L
[ < M
4
or [A
1
/E
R
[ < M
4
.
Otherwise, suppose

X
1
,

X
2
, . . . ,

X
M41
)

U
i
: i A
1
is a sequence of pairwise E
L
-nonequivalent
representatives and that

Y
1
,

Y
2
, . . . ,

Y
M41
)

U
i
: i A
1
are pairwise E
R
-nonequivalent. By ()
we know that for every i, j < M
4
there is some h(i, j) < i

with

X
i

Y
j


U
h(i,j)
. Dene a colouring
f: [M
4
]
3
0, 1, . . . , 6 by:
f(i, j, k) =
_

_
0 if h(i, i) = h(j, k)
1 if h(i, i) = h(k, j)
2 if h(j, j) = h(i, k)
3 if h(j, j) = h(k, i)
4 if h(k, k) = h(i, j)
5 if h(k, k) = h(j, i)
6 otherwise.
(If more then one of these cases occurs, f takes the minimal value.)
By the denition of M
4
there is B M
4
with [B[ = M
3
such that B is homogeneous with respect
to f and we let f
B
m. Is it possible that m < 6? Suppose for example that m = 0, and choose
i < j < j
1
< k from B. If f(i, j, k) = 0 = f(i, j
1
, k) we have h(i, i) = h(j, k) and h(i, i) = h(j
1
, k).
Hence

X
i

Y
i


X
j

Y
k
and

X
i

Y
i


X
j1

Y
k
. It follows that

X
j

Y
k


X
j1

Y
k
and hence
12
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3

X
j
E
L

X
j1
and this is impossible. The other ve possibilities are eliminated similarly and we conclude
that
f
B
6.
Let A
2
:= l < i

: (i B)(h(i, i) = l) and A
3
:= l < i

: (i ,= j B)(h(i, j) = l). By
the choice of B and the above we have A
2
A
3
= . Note that [A
2
[ [B[ = M
3
< K and
[A
3
[ [B[
2
= (M
3
)
2
< K. Hence by the K-randomness of ( there is some k < i

such that
_
l A
2
C [= R(

U
k
,

U
l
,

W)

&
_
l A
3
C [= R(

U
k
,

U
l
,

W)

that is (as R respects )

() i ,= j B C [= [R(

U
k
,

X
i

Y
i
,

W) & R(

U
k
,

X
i

Y
j
,

W)].
By the denition of M
3
= [B[ we have i ,= j B with
() Th
n
(R;

U
k
,

Y
i
,

W) = Th
n
(R;

U
k
,

Y
j
,

W).
But Th
n
(C;

U
k
,

X
i

Y
i
,

W) = Th
n
(L;

U
k
,

X
i
,

W) + Th
n
(R;

U
k
,

Y
i
,

W) =
by ()
Th
n
(L;

U
k
,

X
i
,

W) + Th
n
(R;

U
k
,

Y
j
,

W) = Th
n
(C;

U
k
,

X
i

Y
j
,

W). Therefore:
C [= R(

U
k
,

X
i

Y
i
,

W) C [= R(

U
k
,

X
i

Y
j
,

W)
and this is a contradiction to (), so () is proved.
To conclude, assume [A
1
/E
L
[ < M
4
. Then, by 3.3 and as u
i
: i A
1
is (K,
K
(M2)
2
)-big, there
is A A
1
such that u
i
: i A) is (K +
K
(M2)
2
,
K
(M2)
2
M4
)-big and such that for every i, j A,

U
i
E
L

U
j
. Fix k

A, and dene a sequence

V
i
: i A) by:

V
i

L
=

U

k

L
and

V
i

R
=

U
i

R
.
We want to show that for every i A we have

V
i


U
i
. Indeed, as

U

k
E
L

U
i
and by () we know
that for every r A
1
,

U
k

U
r


U
i

U
r
and choosing r = i we get

U
k

U
i


U
i

U
i
i.e.

V
i


U
i
.
Hence v
i
: i A) is (K
1
, K
2
)-big and all the

V
i
s coincide outside R. Hence R is (K
1
, K
2
)-major.
By a similar argument we get: [A
1
/E
R
[ < M
4
implies L is (K
1
, K
2
)-major.

Notation. K
1
and K
2
will be from now on the numbers from lemma 3.9 above.
The computations below will be useful the in following stages. For the moment assume that (
is nite.
Let (L, R) be a Dedekind cut of C and K > (M
3
)
2
as before. First note that as ( is K-random
we have
[([ = #(C) 2
2(K1)
13
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


By 3.9 we may assume that L is (K
1
, K
2
)-major where K
1
= K +
K
(M2)
2
and K
2
=
K
(M2)
2
M4
. (The
case R is (K
1
, K
2
)-major is symmetric.) If K is big enough we get
2
2(K1)
K
1
= 2
2(K1)
(K +
K
(M
2
)
2
) > K > K
2
=
K
(M
2
)
2
M
4
and by the denition of (K
1
, K
2
)-major
#(L) 2
K2
= 2
K
(M
2
)
2
M
4
.
By 3.7 R is not M
1
(K
1
+K
2
) + 1-fat, i.e.,
#(R) M
1
(K
1
+K
2
) = M
1
(K +
K
(M
2
)
2
+
K
(M
2
)
2
M
4
) 2M
1
K
it follows that
() #(L)/(#(R) + 1)
2
2
K2
/(4(M
1
)
2
K
2
+ 4(M
1
)K + 1) = 2
K
(M
2
)
2
M
4
/(4(M
1
)
2
K
2
+ 4(M
1
)K + 1).
Conclusion 3.10. For every l < there is K

= K

(l, n, d) < such that under the context

in 3.1, if
K K

,
(L, R) is a Dedekind cut of C,
M = M(K, n, d) denotes the bouquet size of the major segment,
m = m(K, n, d) denotes the bouquet size of other segment,
then M/(m+ 1)
2
> l K
2
.
Proof. By the inequality () above and noting that M
1
, M
2
, M
3
and M
4
do not depend on
K.

Remark. By 3.7, if K is big enough then the segment that is not (K

1
, K
2
)-major is minor.
We will always assume that.
If we assume that the interpreted graph ( is innite then we can say that, if K is big enough,
one segment will have an innite bouquet size while the other will have an a priori bounded bouquet
size.
Lemma 3.11. For every n, d < there is K

= K

(n, d) < such that if 1, of dimension d

and depth n, is an interpretation of an innite K-random graph ( on C and K > K

,
then there is m < , that depends only on K, n and d, such that if (L, R) is a Dedekind cut of C:
L (or R) has an innite bouquet size,
R (or L) has bouquet size that is at most m.
14
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


Proof. By lemma 3.9 (letting K

= (M
3
)
2
) we get that one of the segments is (K
1
, K
2
)-major
and hence has innite bouquet size. From 3.7 we get that the other segment is not K
3
-fat where
K
3
depends only on K and the interpretation 1 (i.e. n and d). The required m is that K
3
, which
is good for every Dedekind cut.

4. SemiHomogeneous subsets
Our next step towards reaching a contradiction is of a combinatorial nature. In this section we
introduce the notion of a semi-homogeneous subset and show that the gap between the size of a set
and the size of a semi-homogeneous subset is reasonable.
Denition 4.1. Let K, c < , let I be an ordered set and f: [I]
2
0, . . . , c.
1) We call T I right semi-homogeneous in I (for f and K) if for every i < i

from T we have
[j I : j > i, f(i, j) = f(i, i

)[ K.
2) We call T I left semi-homogeneous in I (for f and K) if for every i < i

from T we have
[j I : j < i

, f(j, i

) = f(i, i

)[ K.
3) We call T I semi-homogeneous in I (for f and K) if T is both right semi-homogeneous
and left semi-homogeneous.
4) We call T I rightnice [leftnice] for S I (and for f and K) if
Max(S)<Min(T) [Min(S)>Max(T)] and for every j T, S j is right semi-homogeneous [left
semi-homogeneous] in T S.
Lemma 4.2. Let K, c, I, f be as above. Suppose [I[ > c N K. Then, there is a right
semi-homogeneous subset S I of cardinality N.
Proof. Let i
0
be Min(I) and T
0
I be of cardinality [I[ c (K 1) such that T
0
is
rightnice for i
0
, (just throw out every j I such that f(i
0
, j) occurs less than K times, there
being at most c (K1) such js). Let i
1
be Min(T
0
) and T
1
T
0
be rightnice for i
1
of cardinality
[I[ 2c (K 1), (use the same argument). Dene S
1
:= i
0
, i
1
. Clearly, for every j T
1
,
S
1
j is right semi-homogeneous in I.
Proceed to dene i
2
(=Min(T
1
) ), T
2
, S
2
and so on. After dening S
N2
and T
N2
we have
thrown out (N 1) c (K 1) elements and as [I[ > c N K we can dene T
N1
, i
N1
and S
N1
which is the required right semi-homogeneous subset.

Lemma 4.3. Let K, c, I, f be as above. Suppose [I[ > c

2
N K
2
= c (cNK) K. Then,
there is a semi-homogeneous subset T I of cardinality N.
Proof. Repeat the construction in the previous lemma to get T

I, right semi-homogeneous
in I of cardinality c N K and now take T T

left semi-homogeneous in T of cardinality N.

T is semi-homogeneous in I.

15
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


We return now to the previous section, and its context. Recall that, given a cut (L, R) we
denoted by M = M(K, n, d) the bouquet size of the major segment and by m = m(K, n, d) the
bouquet size of the minor segment.
Conclusion 4.4. In the context 3.1, for every c, N < there is K < such that if [I[
M(K, n, d) then for every f: [I]
2
0, . . . , c there is a semi-homogeneous subset of T I, for f
and m(K, n, d) + 1, with [T[ N.
Proof. By lemma 4.3. we just need to ensure that c
2
N (m(K, n, d) +1)
2
< M(K, n, d). i.e.
M/(m+ 1)
2
> c
2
N. This holds by conclusion 3.10.

5. The forcing
The universe V
P
where no uniform interpretation exists is the same as in [LiSh]. The forcing
P adds generic semiclubs to each regular cardinal >
0
.
Context. V [= GCH
Denition 5.1. Let >
0
be a regular cardinal
1) SC

:=
_
f : f: 0, 1, < , cf()
0
_
where each f, considered to be a subset of
(or ), is a semiclub. The order is inclusion. (So SC

adds a generic semiclub to ).

2) Q

will be an iteration of the forcing SC

with length
+
and with support < .
3) P := P

, Q

: a cardinal >
0
) where Q

is forced to be Q

if is regular, otherwise it is
. The support of P is Eastons: each condition p P is a function from the class of cardinals
to names of conditions where the class S of cardinals that are matched to non-trivial names is
a set. Moreover, when is an inaccessible cardinal, S has cardinality < .
4) P
<
, P
>
, P

are dened naturally. For example P

<
is P

, Q

:
0
< < ).
Discussion 5.2. Assuming GCH it is standard to see that Q

satises the
+
chain condition
and that Q

and P

do not add subsets of with cardinality < . Hence, P does not collapse
cardinals and does not change conalities, so V and V
P
have the same regular cardinals.
Moreover, for a regular >
0
we can split the forcing into 3 parts, P = P
0
P
1
P
2
where P
0
is P
<
, P
1
is a P
0
-name of the forcing Q

and P
2
is a P
0
P
1
-name of the forcing P
>
such that V
P
and V
P0P1
have the same H(
+
).
In the next sections, when we restrict ourselves to H(
+
) it will suce to look only in V
P0P1
.
6. The contradiction (reduced case)
Collecting the results from the previous sections we will reach a contradiction from the assump-
tion that (for a suciently large K), the monadic theory of some chain C in V
P
, interprets a radom
graph (
K
.
16
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


As we saw in section 3, an interpretation has a major segment. We will show below that there is
a minimal one (and without loss of generality the segment is an initial segment). In this section we
restrict ourselves to a special case: we assume that the minimal major initial segment is the whole
chain C. Moreover, the chain C is assumed to be regular cardinal >
0
.
In the next section we will dispose of these special assumptions. However, the skeleton of those
proofs will be the same as in this reduced case.
Denition 6.1. Assume that (C, <) interprets (
K
by 1. D C is a minimal (K
1
, K
2
)-
major initial segment for 1 if D is an initial segment of C which is a (K
1
, K
2
)-major segment and
no proper initial segment D

D is (K
1
, K
2
)-major.
Fact 6.2. Suppose that (C, <) interprets (
K
by 1, where K and 1 satisfy the assumption
of lemma 3.9. Then there is a chain (C

, <

) that interprets ( by some 1

having the same dimension

and depth as 1, such that there is D

which is a minimal (K
1
, K
2
)-major initial segment for
1

. (K
1
and K
2
are as in lemma 3.9).
Proof. (By [Gu] lemma 8.2). Let L be the union of all the initial segments of C that are
(K
1
, K
2
)-minor (note that if L is minor and L

L then L

is minor as well). If L is (K
1
, K
2
)-major
then set D = L, C

= C, 1

= 1 and we are done.

Otherwise, let D = C L, by lemma 3.9 D is major. Now if there is a proper nal segment
D

D which is (K
1
, K
2
)-major then C D

is minor. But (C D

) L, so that is impossible by
maximality of L. Therefore D is a minimal (K
1
, K
2
)-major (nal) segment. Now take C

to be the
inverse chain of C. Clearly D is a minimal (K
1
, K
2
)-major initial segment for an interpretation 1

of ( (that is obtained be replacing < by > in 1) having the same depth and dimension.

Sketch of the proof. Fixing an interpretation 1 (rather its depth and dimension) we are
trying to show that if K is large enough then in V
P
no chain C interprets some (
K
by 1.
Towards a contradiction we choose K such that

K > N
0
> N
1
> N
2
> N
3
> N
4
> N
5
> N
6
with:
(1) N
6
= max2, n
1
, n
2
, n
3
+ 1 (n
1
, n
2
, n
3
are dened in assumption 5 below).
(2) N
5
(N
6
)
3
32
i.e. a set of size N
5
has a homogeneous subset of size N
6
for colouring triplets
into 32 colours (exists by Ramsey theorem).
(3) N
4
= n
1
N
5
.
(4) N
3
= 2 N
4
.
(5) N
2
(N
3
)
2
n3
(exists by Ramsey theorem).
(6) N
1
(N
2
)
3
32
(exists by Ramsey theorem).
(7) N
0
= n
1
N
1
.
17
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


We start with a sequence

U
i
: i < M) of representatives for the elements of ( (M = #(D) i.e.
the bouquet size of a minimal major segment, in our case it is #(C) = [([, possibly innite), and
gradually reduce their number until we get pairs that will satisfy (for a suitable semiclub a and a
club J):
i < j [

U
i
,

U
j
]
J
a


U
i
.
These will be achieved at steps 1, 2, 3. In steps 4, 5 we will get also:
[

U
i
,

U
j
]
J
a
[

U
j
,

U
i
]
J
a
.
Contradiction will be achieved when we show that some [

U
i
,

U
j
]
J
a
represents two dierent elements.
Assumptions. Our assumptions towards a contradiction are as follows:
1. (C, <) V
P
interprets (
K
by 1 = U(

X,

W), E(

X,

Y ,

W), R(

X,

Y ,

W)),
lg(

X), lg(

Y ) and w.l..o.g lg(

W) = d, n(1) = n.
2. C itself is the minimal (K
1
, K
2
)-major initial segment for 1. Moreover, C = , a regular cardinal
>
0
. For every proper initial segment D C we have #(D) < K
3
. (K
1
, K
2
and K
3
are from 3,
they depend only on K, n and d).
3. m() = m()(n +d, 4d) is as in the preservation theorem 1.11.
4. J =
i
: i < ) is an m()suitable club for all the representatives that will be shued
(there are only nitely many). a is a semiclub, generic with respect to every relevant element
including J (again, nitely many), and see a remark later on.
5. n
1
, n
2
and n
3
are dened as the number of possibilities for the following theories (m() is as
above):
n
1
:= [
_
a-Th
m()
(C;

X,

Y ) :

X,

Y C, lg(

X), lg(

Y ) = d
_
[
n
2
:= [
_
a-Th
m()
(C;

X,

Y ,

Z) :

X,

Y ,

Z C, lg(

X), lg(

Y ), lg(

Z) = d
_
[
n
3
:= [
_
a-Th
m()
(C;

X,

Y ,

Z,

U) :

X,

Y ,

Z,

U C, lg(

X), lg(

Y ), lg(

Z), lg(

U) = d
_
[
6.

K > N
0
. In addition, K is large for l := (n
2
)
2
N
0
(2[T
n,3d
[ + 1)
2
as in conclusion 3.10 i.e.
M/(m+ 1)
2
> l K
2
(this is possible as l depends only on n(1) and d(1)).
To get started we need another observation that does not depend on the special assumption on the
minimal major segment.
Denition 6.3. Suppose D is the minimal (K
1
, K
2
)-major initial segment for the interpreta-
tion. The vicinity of a representative

X denoted by [

X] is the collection of representatives

Y : some

Z

Y coincides with

X outside some proper (hence minor) initial segment of D.
Lemma 6.4. (1) Every vicinity [

X] is the union of at most m = m(K, n, d) (the bouquet
size of a minor segment) dierent equivalence classes.
(2) From Th
n+d
(D;

U
1
,

U
2
,

W) we can compute the truth value of:

U
1
is in the vicinity of

U
2
.
18
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


Proof. If (1) does not hold then there is a proper initial segment D

of the minimal major

initial segment D with #(D

) > m which is impossible. (2) is clear.

We are ready now for a contradiction:

STEP 1: Let N
0
, . . . , N
6
be as above and let K be as in assumption 6.
Let

U
i
: i < M), be a list of representatives for the elements of (
K
that is interpreted by 1 on
C. Let f be a colouring of [M]
2
into n
2
colours dened by
f(i, j) := a-Th
m()
(C;

U
i
,

U
j
,

W).
We would like to get a semi-homogeneous subset of M for f and m+1 of size N
0
. If ( is nite
then this is possible by assumption 6 and conclusion 4.4. Of course if ( is innite (i.e. M
0
) we
can even get a homogeneous one.
Let then S

0, . . . , M 1 be semi-homogeneous and look at B

:=

U
i
: i S

). As
N
0
= n
1
N
1
we can choose
B :=

U
i
: i S)
such that S S

is of size [N
1
[ and such that a-Th
m()
(C;

U
i
,

W) is constant for every i S.
STEP 2: We start shuing the members of B along a and J. Note that by the choice of B
and m() and by the preservation theorem
i, j S Th
n
(C;

U
i
,

W) = Th
n
(C; [

U
i
,

U
j
]
J
a
,

W).
It follows that the results of the shuings are representatives as well, that is
i, j S C [= U([

U
i
,

U
j
]
J
a
,

W).
Dene for i < j S
k(i, j) := min
_
k :
_
k S & [

U
i
,

U
j
]
J
a


U
k
_

_
k = M
_
_
By the choice of N
1
there is a subset A S, of size N
2
, such that for every

U
i
,

U
j
,

U
l
with i < j < l
and i, j, l A, the following ve statements have a constant truth value:
k(j, l) = i,
k(i, l) = j,
k(i, j) = i,
k(i, j) = j,
k(i, j) = l.
Moreover, if there is a pair i < j in A such that k(i, j) A then:
either for every i < j from A, k(i, j) = i or for every i < j from A, k(i, j) = j.
19
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


The reason is the following: suppose that k(, ) = A for some < from A. If < then
k(j, l) = i for all i < j < l from A but k is one valued. Similarly, the possibilities < < and
< are ruled out. We are left with = or = and apply homogeneity.
STEP 3: The aim now is to nd a pair i < j from A with k(i, j) A. Dene A

to be the
results of the shuings:
A

:=
_
k : (i < j A)
_
[

U
i
,

U
j
]
J
a


U
k
__
and it is enough to show that A

A ,= .
If not, as [A[ = N
2
< K and [A

[ [A[
2
< K (we chose

K > N
0
), there is a representative

V
A
such that

iA
[C [= R(

U
i
,

V
A
,

W)]

iA

\A
[C [= R(

U
i
,

V
A
,

W)].
As N
2
> n
2
there is i < j A with:
a-Th
m()
(C;

U
i
,

V
A
,

W) = a-Th
m()
(C;

U
j
,

V
A
,

W)
and by the preservation theorem
() Th
n
(C; [

U
i
,

U
j
]
J
a
,

V
A
,

W) = Th
n
(C;

U
i
,

V
A
,

W).
Now, [

U
i
,

U
j
]
J
a


U
k
for some k A

but by ()
C [= R([

U
k
,

V
A
,

W).
Therefore, by the choice of

V
A
, we have k A. It follows that k A A

so A

A ,= after all.
The aim is fullled and we may assume w.l.o.g that
i < j A [

U
i
,

U
j
]
J
a


U
i
.
STEP 4: The aim now is to show that
i < j A [

U
i
,

U
j
]
J
a
[

U
j
,

U
i
]
J
a
_
= [

U
i
,

U
j
]
J
\a
_
.
Returning to the discussion in 5, we have mentioned so far only a nite number of elements from
H(
+
)
V
P
, (including J). Everything already belongs to H(
+
)
V
P
0
P
1
where P
0
is P
<
and P
1
is
a P
0
-name for Q

which is an iteration of length

+
with support < (we assume that the ground
universe V satises GCH). Moreover, an initial segment of P
0
P
1
, denoted by P
0
P
1

adds all
the relevant elements and we can choose the semiclub a as the one that is generated in the th
stage of P
1
.
Let p P
0
P
1
be a condition that forces all the statements about the representatives we mentioned
so far (e.g. i < j A [

U
i
,

U
j
]
J
a


U
i
). We think about p as a function with domain 1
+
such that p(1) P
0
and for
+
, p() SC

. under this notation p() is an initial segment

20
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


of a and w.l.o.g a member of V
P0
(and not a name for one). Let

= Dom(p()). We may assume

that cf(

) =
0
. Let :=

J (so cf() =
0
as well).
By homogeneity of the forcing, b := (a )
_
( a) [, )

is a semiclub of that is also

generic with respect to the relevant elements. We denote from now on, for

U,

V ,

V :=
_

U
_

V [, )
_
.
For proving we will show that:
() [

U
i
,

U
j
]
J
b


U
i
for all i < j from A,
() [

U
i
,

U
j
]
J
\a

U
k


U
k
for all i, j, k from A,
() [

U
i
,

U
j
]
J
\a

U
i
[

U
i
,

U
j
]
J
\a
for all i < j from A.
STEP 5: Lets prove the claims:
(): By homogeneity of the forcing everything that p forces for a it forces for b.
(): Recall that for every i, j, k A we have
a-Th
m()
(C;

U
i
,

W) = a-Th
m()
(C;

U
j
,

W) = a-Th
m()
(C;

U
k
,

W).
As m() = m()(n+d, 4d) and J satises cf() =
0
we have by the second part of preservation
theorem 1.11
Th
n+d
(C; [

U
i
,

U
j
]
J
a
,

W)
[0,)
= Th
n+d
(C;

U
i
,

W)
[0,)
= Th
n+d
(C;

U
j
,

W)
[0,)
.
Similarly
Th
n+d
(C;

U
i
,

W)
[0,)
= Th
n+d
(C;

U
k
,

W)
[0,)
and it follows that for every i, j, k A:
() Th
n+d
(C; [

U
i
,

U
j
]
J
a
,

W)
[0,)
= Th
n+d
(C;

U
k
,

W)
[0,)
.
Now by the composition theorem
Th
n+d
(C; [

U
i
,

U
j
]
J
a

U
k
,

W) = Th
n+d
(C; [

U
i
,

U
j
]
J
a
,

W)
[0,)
+Th
n+d
(C;

U
k
,

W)
[,)
and this equals by ()
Th
n+d
(C;

U
k
,

W)
[0,)
+Th
n+d
(C;

U
k
,

W)
[,)
= Th
n+d
(C;

U
k
,

W).
As the theories are equal and as

U
k
is a representative, there is some l < M (not necessarily in A)
such that
[

U
i
,

U
j
]
J
a

U
k


U
l
.
If l = k everything is ne. Otherwise assume l > k (symmetrically for l < k) for a contradiction.
21
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


By the denition of vicinity we see that

U
l
[

U
k
] and this is reected in Th
n+d
(C;

U
k
,

U
l
,

W).
Now k A S and S was chosen to be semi-homogeneous in M. Therefore there are l
0
< l
1
<
. . . < l
m
< M with

i<(m+1)
Th
n+d
(C;

U
k
,

U
li
,

W) = Th
n+d
(C;

U
k
,

U
l
,

W).
Hence

i<(m+1)
_
U
li
[U
k
]
_
but by 6.4 a vicinity contains at most m pairwise nonequivalent representatives, a contradiction.
We conclude that l = k.
Therefore, for every i, j, k from A we have [

U
i
,

U
j
]
J
a

U
k


U
k
. Substituting i and j we get: for
every i, j, k from A, [

U
j
,

U
i
]
J
a

U
k


U
k
or:
[

U
i
,

U
j
]
\a

U
k


U
k
.
This is claim ().
(): Now suppose i < j are from A. By denition, for every

P C the theory ( a)-Th
m()
(C;

P)
determines (and is determined by) a-Th
m()
(C;

P). Therefore,
a-Th
m()
(C;

U
i
,

W) = a-Th
m()
(C;

U
j
,

W) & ( a)-Th
m()
(C;

U
i
,

W) = ( a)-Th
m()
(C;

U
j
,

W).
Applying the preservation theorem for a and a we get
Th
n+d
(C; [

U
i
,

U
j
]
J
a
,

W)
[0,)
= Th
n+d
(C;

U
i
,

W)
[0,)
= Th
n+d
(C;

U
j
,

W)
[0,)
and
Th
n+d
(C; [

U
i
,

U
j
]
J
\a
,

W)
[0,)
= Th
n+d
(C;

U
i
,

W)
[0,)
= Th
n+d
(C;

U
j
,

W)
[0,)
so
Th
n+d
(C; [

U
i
,

U
j
]
J
a
,

W)
[0,)
= Th
n+d
(C; [

U
i
,

U
j
]
J
\a
,

W)
[0,)
.
Therefore, as Th
n+d
(C;

P) determines Th
n+d
(C;

P,

P):
() Th
n+d
(C; [

U
i
,

U
j
]
J
a
, [

U
i
,

U
j
]
J
a
,

W)
[0,)
= Th
n+d
(C; [

U
i
,

U
j
]
J
\a
, [

U
i
,

U
j
]
J
\a
,

W)
[0,)
.
By () and () we know that
[

U
i
,

U
j
]
b


U
i
[

U
i
,

U
j
]
J
\a

U
i
and the equivalence is reected by Th
n
(C; [

U
i
,

U
j
]
J
\a

U
i
, [

U
i
,

U
j
]
b
,

W). Clearly Th
n
is determined
by Th
n+d
. Hence:
Th
n
(C; [

U
i
,

U
j
]
J
a

U
i
, [

U
i
,

U
j
]
J
b
) = (by ())
Th
n
(C; [

U
i
,

U
j
]
J
a
, [

U
i
,

U
j
]
J
a
,

W)
[0,)
+Th
n
(C;

U
i
, [

U
i
,

U
j
]
J
\a
,

W)
[,)
=
Th
n
(C; [

U
i
,

U
j
]
\a
, [

U
i
,

U
j
]
\a
,

W)
[0,)
+Th
n
(C;

U
i
, [

U
i
,

U
j
]
\a
,

W)
[,)
=
22
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


Th
n
(C; [

U
i
,

U
j
]
\a

U
i
, [

U
i
,

U
j
]
\a
).
Therefore
[

U
i
,

U
j
]
J
\a

U
i
[

U
i
,

U
j
]
J
\a
and () is proved.
From () and () we conclude
[

U
i
,

U
j
]
J
\a


U
i
[

U
i
,

U
j
]
J
a
and is proved.
STEP 6: By denition of N
2
= [A[, N
3
and N
4
there is a sub-sequence of

U
i
: i A) that will
be denoted for convenience (while preserving the order between the indices) by

P
i
: i < N
3
= 2N
4
)
such that for every i < j < 2N
4
and r < l < 2N
4
:
(i) a-Th
m()
(C;

P
i
,

P
j
,

W) = a-Th
m()
(C;

P
r
,

P
l
,

W) (by dening a colouring of pairs from N
2
).
(ii) [

P
i
,

P
j
]
J
a
[

P
i
,

P
j
]
\a


P
i
(by steps 3 and 5).
For i < N
4
let

Q
i
a representative that satises

[i,2N4i)
_
C [= R(

P

,

Q
i
,

W)
_

[0,i)[2N4i,2N4)
_
C [= R(

P

,

Q
i
,

W)
_
.
As N
4
is big enough there is T 0, . . . , N
4
1 with [T[ = N
6
such that if i, j T then either
[

Q
i
,

Q
j
]
J
a


Q
i
or [

Q
i
,

Q
j
]
J
a


Q
j
. To get T repeat steps 1, 2 and 3 while substituting

U
i
: i < S

)
by

Q
i
: i < N
4
), and N
0
, N
1
, N
2
by N
4
, N
5
and N
6
respectively. Note that we lose generality by
chosing one of the possibilities.
Now choose i, j T (by N
6
> n
3
) such that
a-Th
m()
(C;

P
i
,

P
2N4i
,

Q
i
,

W) = a-Th
m()
(C;

P
j
,

P
2N4j
,

Q
j
,

W)
and shue along a and J:
Th
n
(C;

P
i
,

P
2N4i
,

Q
i
,

W) =
Th
n
(C; [

P
i
,

P
j
]
J
a
, [

P
2N4i
,

P
2N4j
]
J
a
, [

Q
i
,

Q
j
]
J
a
,

W) =
Th
n
(C; [

P
i
,

P
j
]
J
a
, [

P
2N4j
,

P
2N4i
]
J
\a
, [

Q
i
,

Q
j
]
J
a
,

W)
but [

P
i
,

P
j
]
J
a


P
i
, and by step 5,
[

P
2N4j
,

P
2N4i
]
J
\a


P
2N4j
.
Now from [

Q
i
,

Q
j
]
J
a


Q
i
or [

Q
i
,

Q
j
]
J
a


Q
j
and the equality of the theories Th
n
:
C [=
_
R(

P
i
,

Q
i
,

W) & R(

P
2N4j
,

Q
i
)
_
or
C [=
_
R(

P
i
,

Q
j
,

W) & R(

P
2N4j
,

Q
j
).
Both possibilities contradict the choice of the

Q
i
s !
23
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


First Remark. So J and a are chosen as follows: getting

U
i
: i S) at step 1 ([S[ = N
1
)
choose for every subset A S a representative

V
A
that separates

U
i
: i A) from

U
i
: i S A)
(some of these will be the

Q
i
s from step 6). J is an m()-suitable for all these elements and a is
a semiclub that is generic with respect to all of these. Clearly, only nitely many elements are
involved.
Second Remark. Note that genericity was used only at stages 4 and 5 (i.e. to prove
[

U
i
,

U
j
]
J
a
[

U
j
,

U
i
]
J
a
).
We proved the following:
Theorem 6.5. Let U(

X,

Z), E(

X,

Y ,

Z), R(

X,

Y ,

Z)) be a sequence of formulas of dimension
d and depth n.
Then there is K < , that depends only on d and n such that, in V
P
, for no chain C and parameters

W C:
(i) C is isomorphic to a regular cardinal >
0
,
(ii) 1 = U(

X,

W), E(

X,

Y ,

W), R(

X,

Y ,

W)) is an interpretation for some (
k
in C,
(iii) C is the minimal (K
1
, K
2
)-major initial (or nal) segment for 1.

7. Generality
Our aim in this section is to achieve full generality of the interpreting chain C and its minimal
initial major segment D. There are three stages:
(I) D C, D ,= C but D is (isomorphic to) a regular cardinal >
0
.
(II) D = C, D general.
(III) C and D are general.
Let us just remark that always cf(D) >
0
, otherwise we can prove the non existence of interpreta-
tions even from ZFC.
We will elaborate on stages (I) and (II), stage (III) is a simple combination of the techniques.
Chopping o the nal segment. We are trying now to get a contradiction from the same
assumptions as in the previous section except for the following: the minimal (K
1
, K
2
)-major initial
segment D that is a regular cardinal is not necessarily equal to the interpreting chain C. a and J
are therefore subsets of D.
The basic idea of the proof is that if t

is xed and known in advance then to know t

i
+t

all
we need to know is t
i
. Here t
i
are the restrictions of the information (partial theories) to D and t

is
the restriction to C D which can be assumd to be xed, as many representatives coincide outside
D.
24
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


We do not specify the exact size of K (which should be slightly bigger than in the previous
case). It should be apparent however that K depends on n and d only and is obtained by repeated
applications of the Ramsey functions.
Preliminary Step: Let

U
i
: i < [([) a list of representatives for the elements of (. By
denition of D, we may assume that

U
i
: i < M = #(D)) is a list of representatives for a (K
1
, K
2
)-
major subset of ( and all of them coincide outside D. Denote D
@
:= C D and for i < M:

i
:=

U
i
D,

U
@
:=

U
i
D
@
,

:=

W D,

W
@
:=

W D
@
.
Denition 7.1.
(1) Dene on T(D) a unary relation U

(

X) and binary relations

X

Y and R

(

X,

Y ), with
arity d by:
U

(

X) C [= U(

X

U
@
,

W),

Y C [= E(

X

U
@
,

Y

U
@
,

W),
R

(

X,

Y ) C [= R(

X

U
@
,

Y

U
@
,

W).
(When i, j < M for instance, then R

i
,

U

j
) holds if and only if C [= R(

U
i
,

U
j
,

W)).
(2) If i, j < [([ and

U
i
D
@
=

U
j
D
@
we denote
[

U
i
,

U
j
]
J
a
:= [

U
i
D,

U
j
D]
J
a

_

U
i
D
@
_
(If i, j < M for instance then [

U
i
,

U
j
]
J
a
is [

i
,

U

j
]
J
a


U
@
).
Fact 7.2. H(
+
)
V
P
computes correctly

, U

and R

from a-Th
m()
Proof. Take for example

:

X

Y is determined by Th
n
(C;

X

U
@
,

Y

U
@
,

W) =
Th
n
(D;

X,

Y ,

W

) + Th
n
(D
@
;

U
@
,

U
@
,

W
@
).
The second theory is xed for every

X,

Y D. Hence (e.g. by the nite number of possibilities) all
we need to know is the rst theory, which is computed correctly in H(
+
)
V
P
from a-Th
m()
.

We proceed by immitating the previous proof:

STEP 1: Dene B

:=

U
i
: i S

) where S

0, . . . , M 1 is semi-homogeneous , and
B :=

U
i
: i S)
such that S S

, [S[ nite and big enough, with a-Th

m()
(D;

U

i
,

W

) constant for i S.
25
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


STEP 2: Shue the members of B along a and J as in denition 7.1. Note that by the choice
of B and the preservation theorem
i, j S Th
n
(C;

U
i
,

W) = Th
n
(C; [

U
i
,

U
j
]
J
a
,

W)
and therefore the resuts are representatives as well i.e.
i, j S C [= U([

U
i
,

U
j
]
J
a
,

W).
Dene for i < j S
k(i, j) := min
_
k :
_
k S & [

U
i
,

U
j
]
J
a


U
k
_

_
k = M
_
_
equivalently
k(i, j) := min
_
k :
_
k S & [

i
,

U

j
]
J
a

k
_

_
k = M
_
_
Let A S be large enough, homogeneous with the colouring into 32 colours we used before.
STEP 3: The aim is to nd i < j from A with k(i, j) A. Let
A

:=
_
k < [([ : (i < j A)
_
[

U
i
,

U
j
]
J
a


U
k
__
and lets show that A

A ,= .
Othewise, there is some

V
A
(not necessarilly from

U
i
: i < M)) that separates these two disjoint
collections of representatives:

iA
[C [= R(

U
i
,

V
A
,

W)]

iA

\A
[C [= R(

U
i
,

V
A
,

W)].
We may assume that there are i < j from A with
a-Th
m()
(C;

U
i
,

V
A
,

W)
D
= a-Th
m()
(C;

U
j
,

V
A
,

W)
D
.
By the preservation theorem
() Th
n
(C; [

U
i
,

U
j
]
J
a
,

V
A
,

W)
D
= Th
n
(C;

U
i
,

V
A
,

W)
D
and in addition
() Th
n
(C; [

U
i
,

U
j
]
J
a
,

V
A
,

W)
D
@= Th
n
(D
@
;

U
@
,

V
A
D
@
,

W
@
) = Th
n
(C;

U
i
,

V
A
,

W)
D
@ .
Now [

U
i
,

U
j
]
J
a


U
k
for some k A

but by () and () and the composition theorem:

C [= R([

U
i
,

U
j
]
J
a
,

V
A
,

W).
Therefore, by the choice of

V
A
, k A. As k A

it follows that A

A ,= after all.
As before we may conclude that, without loss of generality:
i < j A [

U
i
,

U
j
]
J
a


U
i
26
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


equivalently (and this is known even by H(
+
)
V
P
)
i < j A [

i
,

U

j
]
J
a

i
STEPS 4,5: We work inside H(
+
)
V
P
and concentrate on

i
: i A). The aim is to show
that
i < j A [

i
,

U

j
]
J
a

j
,

U

i
]
J
a
.
Let p P
0
P
1
be a condition that forces all the facts we showed so far about

, U

and R

and
the

U

i
s such as i < j A [

i
,

U

j
]
J
a


U

i
. As before we dene a generic semiclub b and
show that:
() [

i
,

U

j
]
J
b

i
for every i < j from A,
() [

i
,

U

j
]
J
\a

k
for every i, j, k A,
() [

i
,

U

j
]
J
\a

i
,

U

j
]
J
\a
for every i < j from A.
The proofs are exactly the same as in the previous section (substituting C,

W and by D,

W

and

), and from these facts we can deduce . Leaving H(

+
)
V
P
we nd that what we proved in V
P
is:
i < j A [

U
i
,

U
j
]
J
a
[

U
j
,

U
i
]
J
a
STEP 6: As [A[ is big enough we have a sub-sequence of

U
i
: i A) that will be denoted for
convenience (while preserving the order between the indices) by

P
i
: i < N

3
= 2N

4
) such that for
every i < j < 2N

4
and r < l < 2N

4
(N

4
is a suciently big number as usual):
(i) a-Th
m()
(D;

P

i
,

P

j
,

W

) = a-Th
m()
(D;

P

r
,

P

l
,

W

)
(ii) Th
n
(D
@
;

P
@
i
,

P
@
j
,

W
@
) = Th
n
(D
@
;

P
@
r
,

P
@
l
,

W
@
) (

P
i
D
@
is constant),
(iii) [

P
i
,

P
j
]
J
a
[

P
i
,

P
j
]
\a


P
i
.
For i < N

4
let

Q
i
a representative (not necessarily from

U
i
: i < M)) that satises:

[i,2K4i)
_
C [= R(

P

,

Q
i
,

W)
_

[0,i)[2N

4
i,2N

4
)
_
C [= R(

P

,

Q
i
,

W)
_
From the

Q
i
s extract

Q
i
: i T
1
), with [T
1
[ = N

5
large enough such that for every i < j from
T
1
:
_
a-Th
m()
(D,

Q

i
,

W

) = a-Th
m()
(D,

Q

j
,

W

)
_
&
_
Th
n
(D
@
,

Q
@
i
,

W
@
) = Th
n
(D
@
,

Q
@
j
,

W
@
)
_
where

Q

i
:=

Q
i
D and

Q
@
i
:=

Q
i
D
@
.
For i < j in T
1
denote
[

Q
i
,

Q
j
]
J
a
:= [

Q

i
,

Q

j
]
J
a


Q
@
i
and note that by the preservation theorem the results of the shuings are representatives for elements
of ( i.e.
i, j T
1
C [= U([

Q
i
,

Q
j
]
J
a
,

W).
27
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


Dene k

(i, j) for i < j from T

1
by
k

(i, j) := min
_
k :
_
k T
1
& [

Q
i
,

Q
j
]
J
a


Q
k
_

_
k = N

5
_
_
.
There is a subset T T
1
of size N

6
, large enough, such that for all

Q
i
,

Q
j
,

Q
l
with i < j < l from
T the following ve statements have a constant truth value: k

(j, l) = i, k

(i, l) = j, k

(i, j) = i,
k

(i, j) = j, k

(i, j) = l. Moreover, as usual if there are i < j in T with k

(i, j) T then either for

every i < j in T, k

(i, j) = i or for every i < j in T, k

(i, j) = j.
If there isnt such a pair choose

V
T
such that

iT
[C [= R(

Q
i
,

V
T
,

W)]

i<jT
[C [= R([

Q
i
,

Q
j
]
J
a
,

V
T
,

W)]
as N

6
is big enough there are i < j from T with:
a-Th
m()
(D;

Q

i
,

V
T
D,

W

) = a-Th
m()
(D;

Q

j
,

V
T
D,

W

)
Th
n
(D
@
;

Q
@
i
,

V
T
D
@
,

W
@
) = Th
n
(D
@
;

Q
@
j
,

V
T
D
@
,

W
@
)
By the preservation theorem and the composition theorem we get
() Th
n
(C; [

Q
i
,

Q
j
]
J
a
,

V
T
,

W) = Th
n
(C;

Q
i
,

V
T
,

W).
Therefore C [= R([

Q
i
,

Q
j
]
J
a
,

V
T
,

W) which is a contradition.
It follows: either i, j T [

Q
i
,

Q
j
]
J
a


Q
i
or i < j T [

Q
i
,

Q
j
]
J
a


Q
j
.
Now choose i, j T such that
a-Th
m()
(D;

P

i
,

P

2N

4
i
,

Q

i
,

W

) = a-Th
m()
(D;

P

j
,

P

2N

4
j
,

Q

j
,

W

)
Th
n
(D
@
;

P
@
i
,

P
@
2N

4
i
,

Q
@
i
,

W
@
) = Th
n
(D
@
;

P
@
j
,

P
@
2N

4
j
,

Q
@
j
,

W
@
)
Shue along a and J and get a contradiction as before to the denition of the

Q
i
s.
We have proved the following:
Theorem 7.3. Let U(

X,

Z), E(

X,

Y ,

Z), R(

X,

Y ,

Z)) be a sequence of formulas of dimension
d and depth n.
Then there is K < , that depends only on d and n such that, in V
P
, for no chain C and parameters

W C:
(i) 1 = U(

X,

W), E(

X,

Y ,

W), R(

X,

Y ,

W)) is an interpretation for some (
k
in C,
(ii) D, the minimal (K
1
, K
2
)-major initial (or nal) segment for 1, is isomorphic to a regular
cardinal >
0
.

Reduced Shuings: There are two main diculties that face us in the general context.
The rst one is that the preservation theorem is formulated only in the context of well ordered
chains. We can try and solve this by choosing a conal sequence through the chain and shue along
this sequence. However the second diculty is that a semiclub that has the cardinality of cf(D)
(where D is the minimal major initial segment) cant be generic with respect to subsets of D when
28
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


[D[ > cf(D). The solution for both this diculties lies in the observation that what we really shue
are not subsets of the chain but rather partial theories.
Suppose that we are given a chain C, with cf(C) = >
0
and some

A C of length l.
For simplicity we assume that the chains have a rst element min(C). Choosing a conal sequence
E =
i
: i < ) in C such that
0
= min(C) and dening s
i
:= Th
n
(C,

A)
[i,i+1)
we get by the
composition theorem that
Th
n
(C,

A) =

i<
s
i
.
Concentrating on the chain (, <) we dene a sequence

P =

P
A
= P
t
: t T
n,l
) where for t T
n,l
,
P
t
:= i < : s
i
= t. By the Feferman-Vaught theorem (1.9) we know that Th
n
(C;

A) is determined
by Th
m
(;

P) where m = m(n, l) depends only on n and l.
Lemma 7.4. Let C be a chain with conality >
0
and let n, l N. Then, there are
m(), l(), () N, all depending only on n and l, such that
(a) there is a 1-1 function

X

P
X
such that for every

A C of length l there is

P
A
of
length l() and Th
n
(C;

A) is determined by Th
m()
(;

P
A
),
(b) () codes a Turing machine that computes Th
n
(C;

A) from Th
m()
(;

P
A
).
Proof. Choose a conal E =
i
: i < ) (
0
= min(C)). Let

P
A
be as above () = [T
n,l
[
and (a) is clear from the previous discussion. The computability in clause (b) is clear from the fact
that T
m(),l()
and T
n,l
are both nite.

Remark. Of course we dont really lose generality by assuming that C has a minimal element.
If C interprets ( by 1 and doesnt have one then we can always construct C

= C and
interpret ( on C

by some 1

having the same depth and dimension d+1 (add as a parameter).

So instead of taking K = K(n, d) we use K = K(n, d + 1) for getting a contradiction.
The discussions above justify the following denition:
Denition 7.5. Let n, d N, and t
k
: k < [T
n,d
[) be the list of the possibilities T
n,d
.
(1) T = (,

P) is a pre-chain if >
0
is a regular cardinal, and

P = P
k
: k < [T
n,d
[) is a
partition of .
(2) We identify (,

P) with s
i
: i < ) when i P
k
s
i
= t
k
.
(3)
_
(,

P), E
_
is a guess for (C,

A) if:
(i) E =
i
: i < ) is conal in C and
0
= min(C),
(ii)

A C and lg(

A) = d,
(iii) Th
n
(C;

A)
[i,i+1)
= s
i
when (,

P) = s
i
: i < ).
Next we claim that the guesses (which are well ordered chains of the correct cardinality) repre-
sent faithfully the guessed chain.
29
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


Denition 7.6. Suppose that
a. C is a chain with conality >
0
,
b.

A,

B C have length d,
c. E =
i
: i < ) is conal in C and
0
= min(C),
d. J =
j
: j < ) is a club and a a semiclub.
The reduced shuing of

A and

B along E, J and a, denoted by [

A,

B]
J,E
a
is dened by:
[

A,

B]
J,E
a
:=
_
ja
_

A [
j
,
j+1
)
_

_
ja
_

B [
j
,
j+1
)
_
Fact 7.7. If C,

A,

B, J and a are as above,
_
(,

P
A
), E
_
a guess for (C,

A) and
_
(,

P
B
), E
_
a guess for (C,

B) then
[

P
A
,

P
B
]
J
a
=

P
[

A,

B]
J,E
a
Proof. Straightforward.

Denition 7.8. For C,

A C, E C as above and a a semiclub, dene
a-Th
n
E
(C;

A) := a-Th
n
(,

P
A
)
Lemma 7.9. For every n, d N there is k() = k(n, d) N such that if
1. C is a chain and cf(C) = >
0
,
2.

A,

B C are of length d,
3. E is conal in C,
4. a is a semiclub,
then
a-Th
k()
E
(C;

A) = a-Th
k()
E
(C;

B) Th
n
(C;

A) = Th
n
(C;

B) = Th
n
(C; [

A,

B]
J,E
a
).
Proof. Let k() be k
_
m(), [Tn, d

) where m() is m()(n, d) from the preservation theorem,

and k(, ) is the Feferman-Vaught number as in theorem 1.9.

Lemma 7.10. Let U(

X,

Z), E(

X,

Y ,

Z), R(

X,

Y ,

Z)) be a sequence of formulas of dimension
d and depth n.
Then there is K < , that depends only on d and n such that in V
P
, for no chain C and
parameters

W C:
(i) 1 = U(

X,

W), E(

X,

Y ,

W), R(

X,

Y ,

W)) is an interpretation for some (
k
in C,
(ii) D, the minimal (K
1
, K
2
)-major initial (or nal) segment for 1, satises cf(D) = >
0
.
30
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


Proof. We will follow the previous procedures, this time choosing K big enough with respect
to k() as above and not m() as usual. Assume that 1 is an interpretation of (
K
and suppose
rst that C = D. Fix E C conal, of order type . As [E[ = , it belongs to the intermediate
V
P

.
Let

U
i
: i < [([) a list of the representatives and after the preliminary colouring we remain
with a semi-homogeneous list B :=

U
i
: i S), ([S[ = N
1
big enough) having now the same
a-Th
k()
E
(C;

U
i
). Let B
1
=

V
j
: j < [S[
N2
) a list of the representatives for elements separating
subsets of B of size N
2
from their complements.
Let
_
(,

P
,,
, E
_
: , , < N
N2
1
) be a list of all the guesses for chains of the form
(C;

A
0
,

A
1
,

A
2
,

W) with

A
i
B B
1
for i < 3.
Choose J , a k() suitable club for all the guesses, and a generic semiclub a . Start
shuing

U
i
: i S) (i.e. the respective guesses). A statement of the form

U

is translated to
Th
m()
(;

P
U
,

P
U

) is such that C [= E(

,

U

,

W).
Repeating the usual steps we get

U
i
: i A) such that w.l.o.g [

U
i
,

U
j
]
J,E
a


U
i
for every i < j
from A. Using genericity we can show also that [

U
j
,

U
i
]
J,E
a


U
i
as well.
Now choose a sequence of separating representatives

Q
i
: i < [A[/2) from B
1
above (so J is
suitable for them as well) and get a contradiction as usual.
In the case D ,= C we combine the above with the previous proof: the result of the shuing of
a pair of representatives

U

and

U

(coinciding outside D) is:

_
the result of the reduced shuing of

U

D and

U

D
_ _

(C D)
_
..
And we work in D.

As an -random graph is K-random for each K < we proved:

Theorem 7.11. In V
P
:
A. If C
K
, I,

W
K
: K A ) is a uniform interpretation of
n
in the monadic theory of
order and D
K
C
K
are the minimal major initial (or nal) segments of the interpretations, then
cf(D
K
)
0
for every large enough K.
B. If 1 = U(

X,

W), E(

X,

Y ,

W), R(

X,

Y ,

W)) is an interpretation for RG

in a chain C and
D C is the minimal major initial (or nal) segment then cf(D)
0
.

In the next section we will show that cf(D)

0
is impossible even from ZFC.
8. Short Chains
Recall that a short chain is a chain that does not embed (
1
, <) and the inverse chain (
1
, >).
Our aim in this section is to prove, from ZFC, the non exsistence of interpretations in short chains.
31
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


In fact we show (and this is the only possibility when C is short) the non exsistence of interpretations
with cf(D)
0
.
Denition 8.1. An interpretation 1 of (
K
in a chain C is a short interpretation if the
minimal (K
1
, K
2
)-major initial (or nal) segment for 1, has conality
0
.
The case cf(D) <
0
is impossible:
Fact 8.2. Let 1 be an interpretation of some (
K
in C. Let D be the (K
1
, K
2
)-major
initial segment. Then (if K is suciently big with respect to d(1) and n(1)), D does not have a
last element.
Proof. When K is big enough we have M(K, n, d)/m(K, n, d) > 2 (by 3.10) and this what
we need. Now if D = D

x where x is the last element of D then, from the denitions, easily

#(D)/#(D

) 2. But D

is minor and this is a contradiction.

Assumptions. From now on we are assuming towards a contradiction:

1. 1 = U(

X,

W), E(

X,

Y ,

W), R(

X,

Y ,

W) is an interpretation for some (
K
in a chain C.
n(1) = n and d(1) = d,
2. K = K(n, d) is big enough (we will elaborate later),
3. C has a minimal element (almost w.l.o.g by a previous remark),
4. C is the minimal major initial segment for 1,
5. cf(C) =
0
.
The next denition is the current replacement of m()-suitable club:
Denition 8.3. Let

U
i
: i < i

) be with

U
i
C, lg(

U
i
) = d. Let E =
k
: k < ) C be
increasing in C. E is an r-suitable sequence for

U
i
: i < i

) if
1. E is conal in C and
0
= min(C),
2. For every i < j < i

there is t
i,j
T
r,3d
such that for every 0 < k < :
Th
r
(C;

U
i
,

U
j
,

W)
[0,
k
)
= t
i,j
,
3. For every i < j < i

there is s
i,j
T
r,3d
such that for every 0 < k < l < :
Th
r
(C;

U
i
,

U
j
,

W)
[
k
,
l
)
= s
i,j
.
r-suitable sequences exist:
Claim 8.4.
1. Suppose that

U,

V C are of length d and E =
k
: k < ) is r-suitable for

U,

V . Let
E
1
E be innite with
0
E
1
. Then E
1
is r-suitable for

U,

V .
2. Let

U,

V C be as above and let E =
k
: k < ) be conal with
0
= min(C). Then
there is E
1
E that is r-suitable for

U,

V .
3. For every nite family

U
i
: i < i

) with

U
i
C, lg(

U
i
) = d there is an r-suitable E C.
32
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


Proof. The rst part is immediate. For proving 2. let

U,

V , E be given. Let f: [ 0]
2

[T
r,3d
[ [T
r,3d
[ be a colouring dened (for 0 < k < l < ) by
f(k, l) =

Th
r
(C;

U
i
,

U
j
,

W)
[0,
k
)
, Th
r
(C;

U
i
,

U
j
,

W)
[
k
,
l
)
_
Let u be innite, homogeneous with respect to f (T
r,3d
is nite). Dene E
1
:=
0

k
: k
u.
The third part is immediate by 1. and 2.

We will assume

K N
0
N
1
N
2
0, all depending only on n and d.
Let M = [([ and let

U
i
: i < M) be a list of representatives for the elements of (. Let
f: [M]
2
[T
n+d,3d
[ be dened by
f(i, j) = Th
n+d
(C;

U
i
,

U
j
,

W).
We may assume that there is S of size N
0
, semi-homogeneous with respect to f and (m+1), where
m is the bouquet size of minor segments.
Let E =
k
: k < ) C be (n + d)-suitable for

U
i


W : i S

), (by 8.4). Let S S

be of
size N
1
such that for every i < j and r < s from S, and for every 0 < k < l < :
Th
n+d
(C;

U
i
,

U
j
,

W)
[0,
k
)
= Th
n+d
(C;

U
r
,

U
s
,

W)
[0,
k
)
:= t
and
Th
n+d
(C;

U
i
,

U
j
,

W)
[
k
,
l
)
= Th
n+d
(C;

U
r
,

U
s
,

W)
[
k
,
l
)
:= s.
This is possible by the denition of (n+d)-suitability (and as N
0
is big enough). By the composition
theorem for every i < j in S:
Th
n+d
(C;

U
i
,

U
j
,

W) = t +

k<
s .
Denition 8.5. For u dene the shuing of

U
i
and

U
j
along u by
[

U
i
,

U
j
]
u
:=
_
ku
_

U
i
[
k
,
k+1
_

_
ku
_

U
j
[
k
,
k+1
_
Claim 8.6. For every i < j in S, for every u , C [= U([

U
i
,

U
j
]
u
,

W).
Proof. By suitability of E and denition of S there are t
0
and s
0
such that for every i S:
(i) Th
n
(C;

U
i
,

W)
[0,
k
)
= t
0
for every 0 < k < ,
(ii) Th
n
(C;

U
i
,

W)
[
k
,
l
)
= s
0
for every 0 < k < l < ,
(iii) Th
n
(C;

U
i
,

W) = t
0
+

k<
s
0
.
By the denition of shuing, for every u and i < j in S,
Th
n
(C; [

U
i
,

U
j
]
u
,

W) = t
0
+

k<
s
0
= Th
n
(C;

U
i
,

W).
Therefore C [= U([

U
i
,

U
j
]
u
,

W).

33
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


Dene now:
e := 2k : k < ,
o := 2k + 1 : k < ,
p := 0,
q := 0.
Let
k(i, j) := min
_
k :
_
k S & [

U
i
,

U
j
]
e


U
k
_

_
k = [([
_
_
.
By bigness of N
1
there is A S of size N
2
such that for every

U
i
,

U
j
,

U
l
with i < j < l from A, the
following, usual, ve statements have the same truth value:
k(j, l) = i,
k(i, l) = j,
k(i, j) = i,
k(i, j) = j,
k(i, j) = l.
Moreover, (the usual proof) if for some i < j in A, k(i, j) A then: either for every i < j in A,
k(i, j) = i or for every i < j in A, k(i, j) = j.
Lets nd i < j in A with k(i, j) A: if we cant then there is some

V
A
that separates between
/
1
:=

U
i
: i A and /
2
:=

U
l
: (i < j A)([

U
i
,

U
j
]
e


U
l
. i.e.

Ui/1
_
C [= R(

U
i
,

V
A
,

W)
_

Ui/2
_
C [= R(

U
i
,

V
A
,

W)
_
.
We may assume that E is suitable also for

V
A
(there are nitely many possibilities for

V
A
after
choosing

U
i
: i S

)). As N
2
is big enough there are i < j in A such that for every 0 < k < l <
Th
n
(C;

U
i
,

V
A
,

W)
[0,
k
)
= Th
n
(C;

U
j
,

V
A
,

W)
[0,
k
)
and
Th
n
(C;

U
i
,

V
A
,

W)
[
k
,
l
)
= Th
n
(C;

U
j
,

V
A
,

W)
[
k
,
l
)
.
It follows that
Th
n
(C;

U
i
,

V
A
,

W) = Th
n
(C; [

U
i
,

U
j
]
e
,

V
A
,

W)
and /
1
/
2
,= , a contradiction. We conclude,
() (i < j)in A such that [

U
i
,

U
j
]
e


U
i
or [

U
i
,

U
j
]
e


U
j
.
Fact 8.7. For every i, j in A (in fact in S

): [

U
i
,

U
j
]
q


U
j
and [

U
i
,

U
j
]
p


U
i
.
34
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


Proof. Lets prove the rst statement (the second is proved similarly). By claim 8.6, [

U
i
,

U
j
]
q
is a representative hence is equivalent to

U
l
for some l < [([. Suppose that l > j. By semi-
homogeneity of S

(therefore of A) there are j < l

0
< l
1
. . . < l
m+1
such that

r<m+1
Th
n+d
(C;

U
j
,

U
lr
) = Th
n+d
(C;

U
j
,

U
l
).
By denition of q,

U
l
belongs to the vicinity of

U
j
. As belonging to the vicinity is determined
by Th
n+d
we get m+1 pairwise nonequivalent representatives in [

U
j
]. This is impossible by lemma
6.4. The same holds if we assume l < j. Therefore we must conclude l = j i.e. [

U
i
,

U
j
]
q


U
j
.

Returning to the representatives

U
i
and

U
j
we got in () above, suppose rst that [

U
i
,

U
j
]
e


U
i
.
We will show that
(1) [

U
i
,

U
j
]
e


U
i
[

U
i
,

U
j
]
o
[

U
i
,

U
j
]
q
,
(2) [

U
i
,

U
j
]
e
[

U
i
,

U
j
]
o
.
It will follow that [

U
i
,

U
j
]
e
[

U
i
,

U
j
]
q
and by the previous fact

U
i


U
j
which is a contradiction.
For showing (1) it is enough to show that
Th
n
(C;

U
i
, [

U
i
,

U
j
]
e
,

W) = Th
n
(C; [

U
i
,

U
j
]
o
, [

U
i
,

U
j
]
q
,

W).
Remembering how S was chosen we get
Th
n
(C;

U
i
, [

U
i
,

U
j
]
e
,

W)
[0,1)
= Th
n
(C;

U
i
,

U
i
,

W)
[0,1)
=
Th
n
(C; [

U
i
,

U
j
]
o
, [

U
i
,

U
j
]
q
,

W)
[0,1)
.
Th
n
(C;

U
i
, [

U
i
,

U
j
]
e
,

W)
[
2k
,
2k+1
)
= Th
n
(C;

U
i
,

U
i
,

W)
[
2k
,
2k+1
)
=
Th
n
(C;

U
j
,

U
j
,

W)
[
2k
,
2k+1
)
= Th
n
(C; [

U
i
,

U
j
]
o
, [

U
i
,

U
j
]
q
,

W)
[
2k
,
2k+1
)
Th
n
(C;

U
i
, [

U
i
,

U
j
]
e
,

W)
[
2k+1
,
2k+2
)
= Th
n
(C;

U
i
,

U
j
,

W)
[
2k+1
,
2k+2
)
=
Th
n
(C; [

U
i
,

U
j
]
o
, [

U
i
,

U
j
]
q
,

W)
[
2k+1
,
2k+2
)
and Th
n
(C;

U
i
, [

U
i
,

U
j
]
e
,

W) = Th
n
(C; [

U
i
,

U
j
]
o
, [

U
i
,

U
j
]
q
,

W) follows from the composition
theorem.
For (2) note that
Th
n
(C; [

U
i
,

U
j
]
e
,

U
i
,

W) = Th
n
(C; [

U
i
,

U
j
]
e
,

U
i
,

W)
[0,1)
+
Th
n
(C; [

U
i
,

U
j
]
e
,

U
i
,

W)
[1,2)
+Th
n
(C; [

U
i
,

U
j
]
e
,

U
i
,

W)
[2,3)
+. . .
= Th
n
(C;

U
i
,

U
i
,

W)
[0,1)
+Th
n
(C;

U
j
,

U
i
,

W)
[1,2)
+Th
n
(C;

U
i
,

U
i
,

W)
[2,3)
+
Th
n
(C;

U
j
,

U
i
,

W)
[3,4)
+Th
n
(C;

U
i
,

U
i
,

W)
[4,5)
+. . .
and that
Th
n
(C; [

U
i
,

U
j
]
o
,

U
i
,

W) = Th
n
(C; [

U
i
,

U
j
]
o
,

U
i
,

W)
[0,2)
+
35
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


Th
n
(C; [

U
i
,

U
j
]
o
,

U
i
,

W)
[2,3)
+Th
n
(C; [

U
i
,

U
j
]
o
,

U
i
,

W)
[3,4)
+. . .
= Th
n
(C;

U
i
,

U
i
,

W)
[0,2)
+Th
n
(C;

U
j
,

U
i
,

W)
[2,3)
+Th
n
(C;

U
i
,

U
i
,

W)
[3,4)
+
Th
n
(C;

U
j
,

U
i
,

W)
[4,5)
+Th
n
(C;

U
i
,

U
i
,

W)
[5,6)
+. . .
By the composition theorem:
Th
n
(C; [

U
i
,

U
j
]
e
,

U
i
,

W) = Th
n
(C; [

U
i
,

U
j
]
o
,

U
i
,

W).
That is:
[

U
i
,

U
j
]
e


U
i
[

U
i
,

U
j
]
o
.
Collecting the results we get:
[

U
i
,

U
j
]
e


U
i
(this is the assumption),

U
i
[

U
i
,

U
j
]
o
(by (2) above),
[

U
i
,

U
j
]
o
[

U
i
,

U
j
]
q
(by (1) above),
[

U
i
,

U
j
]
q


U
j
(by fact 8.7).
Therefore,

U
i


U
j
a contradiction.
We are therefore forced to assume that [

U
i
,

U
j
]
e


U
j
but then we get the same way

U
i
[

U
i
,

U
j
]
o
(like (2) above), [

U
i
,

U
j
]
o
[

U
i
,

U
j
]
p
(like (1) above), [

U
i
,

U
j
]
p


U
j
(by 8.7), and again

U
i


U
j
.
We assumed that C is equal to D, the minimal major initial segment for simplicity. However,
if D ,= C then following previous procedures we can easily chop o C D and basically work inside
D, getting a contradiction.
So we have eliminated the possibilities that were left by theorem 7.11 and proved:
Theorem 8.8 (Non-Interpretability Theorem). There is a forcing notion P such that in
V
P
the following hold:
(1) RG

is not interpretable in the monadic theory of order.

(2) For every sequence of formulas 1 = U(

X,

Z), E(

X,

Y ,

Z), R(

X,

Y ,

Z)) there is K

< ,
(eectively computable from 1), such that for no chain C,

W C, and K K

does U(

X,

W),
E(

X,

Y ,

W), R(

X,

Y ,

W)) interpret RG
K
in C.
(3) The above propositions are provable in ZFC. if we restrict ourselves to the class of short chains.

36
5
2
7


r
e
v
i
s
i
o
n
:
1
9
9
7
-
0
7
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
7
-
0
7
-
2
3


REFERENCES
[Ba] J. BALDWIN, Denable secondorder quantiers, Model Theoretic Logics, (J. Barwise
and S. Feferman, editors), SpringerVerlag, Berlin 1985, pp. 445477.
[BaSh] J. BALDWIN and S. SHELAH, Classication of theories by second order quantiers, Notre
Dame Journal of Formal Logic, vol. 26 (1985) pp. 229303.
[Gu] Y. GUREVICH, Monadic secondorder theories, Model Theoretic Logics, (J. Barwise
and S. Feferman, editors), SpringerVerlag, Berlin 1985, pp. 479506.
[GMS] Y. GUREVICH, M. MAGIDOR and S. SHELAH, The monadic theory of
2
, The Journal
of Symbolic Logic, vol. 48 (1983) pp. 387398.
[GuSh] Y. GUREVICH and S. SHELAH, On the strength of the interpretation method, The
Journal of Symbolic Logic, vol. 54 (1989) pp. 305323.
[GuSh1] Y. GUREVICH and S. SHELAH, Monadic theory of order and topology in ZFC, Ann.
Math. Logic, vol. 23 (1982) pp. 179182.
[GuSh2] Y. GUREVICH and S. SHELAH, Interpreting secondorder logic in the monadic theory
of order, The Journal of Symbolic Logic, vol. 48 (1983) pp. 816828.
[GuSh3] Y. GUREVICH and S. SHELAH, The monadic theory and the next world, Israel
Journal of Mathematics, vol. 49 (1984) pp. 5568.
[LiSh] S. LIFSCHES and S. SHELAH, Peano arithmetic may not be interpretable in the monadic
theory of linear orders, The Journal of Symbolic Logic, accepted for publication.
[Sh] S. SHELAH, The monadic theory of order, Annals of Mathematics, ser. 2, vol. 102
(1975) pp. 379419.
[Sh1] S. SHELAH, Notes on monadic logic, part B: complexity of linear orders, Israel Journal
of Mathematics, vol. 69 (1990) pp. 99116.
[TMR] A. TARSKI, A. MOSTOWSKI and R. M. ROBINSON, Undecidable theories, North-
Holland Publishing Company, 1953, xi+98 pp.
37