8

6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


EXISTENCE OF EF-EQUIVALENT NON ISOMORPHIC
MODELS
CHANOCH HAVLIN AND SAHARON SHELAH
Abstract. We prove the existence of pairs of models of the same car-
dinality which are very equivalent according to EF games, but not
isomorphic. We continue the paper [4], but we dont rely on it.
Contents
1. Introduction 2
2. Games with trees for regular =

0
3
3. Games with trees for singular =

0
9
4. regular >

13
5. > cf() >

18
References 23
References 23
1991 Mathematics Subject Classication. MSC (2000) 03250.
Key words and phrases. model theory, innary logics, EF equation, non-isomorphic.
We would like to thank the Israel Science Foundation for partial support of this research
(Grant No. 242/03). Publication 866 of the second author.
1
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


2 CHANOCH HAVLIN AND SAHARON SHELAH
1. Introduction
There had been much study of equivalence relations between models. When
we study such an equivalence relation, the basic question is whether this
relation is actually trivial, i.e., if equivalent models are isomorphic. For
example, countable models which are elementary equivalent in L

1
,
are
isomorphic. (Scott showed this in [10] for countable vocabulary, and Chang
generalized it in [1] for any vocabulary). For = cf() >
0
, Morely gave
(without publishing) a counter example - a pair of L
,
equivalent models
of size which are not isomorphic. Shelah [9, Ch. II, 7] gave such an
example for almost every singular .
Those questions also relate to classication theory: The existence of
strongly equivalent models which are not isomorphic is a non-structure
property for a class of models. On the other side, if a not too strong
equivalence relation is actually the isomorphism relation, this is a structure
property.(See [8] and [2]).
One of the equivalence relations studied in this context, is equivalence
under EF(Ehernfeucht-Fraisse) games. A detailed discussion of EF games
and their history can be found in [3] and in [Vaa95]. The general structure
of an EF game on a pair of models is as follows: There are two players -
isomorphism player, who we call ISO and anti-isomorphism player, who we
call AIS. During the game, AIS chooses members of the models, and ISO
denes interactively a partial isomorphism between the models - in every
move he has to extend that partial isomorphism so that the elements chosen
by AIS will be contained in the domain or in the range. The isomorphism
player loses the game if at some point he cannot nd a legal move. If he
does not lose, he wins. We limit the length of the game and the number
of elements that AIS may choose at each move. (Because, if AIS can list
all the members of one of the models, then the game is not interesting).
In [4], the games were with xed length. In this paper, we deal with EF
games approximated by trees - the length of the game is limited by adding
the demand that in each move, AIS has to choose a node in some xed tree
T (with certain properties) such that the sequence of nodes formed by his
choices is strictly increasing in the order <
T
. If AIS cannot choose such
node - he loses.
We say that two models are equivalent with respect to some EF game if
ISO has a winning strategy in played on those models.
In [4] it was proved that if = cf() =

0
, then there are non-isomorphic
models of size which are EF
,
equivalent for every < , where EF
,
equivalence means that they are equivalent under every EF game with
stages, such that AIS has to choose < members of the models at each
stage. There was also a result for singular, with a necessary change of the
equivalence relation.
Here we generalize the results in two ways: First, we move to EF games
approximated by trees instead of xed-length games(see Hyttinen and Tuuri
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


EXISTENCE OF EF-EQUIVALENT NON ISOMORPHIC MODELS 3
in [2] who investigated such games in the context of classication theory).
Second, we give results also for >

without the assumption =

0
,
where we use PCF theory to have some approximation instead of =

0
.
In Section 2 we prove that for regular =

0
for some class of reasonably
large trees (see detailed discussion justifying the choice, in the beginning of
Section 2) for every tree from that class there are non isomorphic models of
size which are equivalent under EF games approximated by that tree such
that in each move AIS is allowed to choose < members of the models (see
Denition 2.1). Sec1-Def-Game-
With-Tree) In Section 3 we do the parallel for singular . But for singular , if we
allow AIS to choose < elements in each move, and the tree has a branch
of length cf(), then the game is not interesting, because AIS can choose
all the members of the models during the game. So we have to be more
careful - we allow AIS to choose only one element in each move. This is
still a generalization of the result for such in [4] - see the discussion at the
beginning of Section 3.
In Section 4 we prove that for regular >

, for every tree of size

without a branch of length there are non-isomorphic models of size which
are equivalent under the EF game approximated by that tree such that in
each move AIS is allowed to choose < members of the models.
In Section 5 we prove a similar result for > cf() >

. As we explained
above, because of the singularity of , we have to restrict the number of
elements that AIS is allowed to choose at each move - in stage , AIS has to
choose < 1 + members of the models. For a further work in preparation
of the second author on this subject see [?] on his web. nisux
2. Games with trees for regular =

0
In [2] there is a construction of non-isomorphic models of size which are
equivalent under EF games approximated by trees of size with no branch,
when =
<
. In [4] there is such a construction under a weaker assumption
on : = cf() =

0
, but there the result is for games of any xed
length < , not for games which are approximated by trees. We want to
generalize this result to games approximated by trees.
Now, which trees should we consider? If we limit ourselves only to trees
of size , it seems that the set of trees will be small. Why? Assume for
example that = cf() =

0
<

1
. A tree of size must drop at least
one of the following conditions:
1. Above every node there is an antichain of size .
2. Every chain of size
1
has an upper bound.
If
1
, this kind of trees seem to be too degenerate. We could have
demanded that the size of the tree will be 2
<
. But it is possible that
2
<
= 2

and it is reasonable to assume that the result will not be true in

this case.
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


4 CHANOCH HAVLIN AND SAHARON SHELAH
We take the middle road: we do not limit explicitly the size of the tree,
but we demand that the tree will be denable enough - the cause of not
having a branch of length is that the nodes of the tree are actually partial
functions from to which satisfy a certain local condition. By local we
mean that a function f satises the condition i any restriction of f to a
countable set satises it. The tree order is inclusion, and there is no function
from to which satises the condition. By Remark 2.4 this result is indeed Sec1-Remark-
Strength-Of-
Game)
a generalization of for every tree of size and without a branch.
LABEL Sec1-Def-
Game-With-Tree)
Denition 2.1. For a tree T , a cardinal , and models with common vo-
cabulary M
1
, M
2
, we dene the game
T ,
(M
1
, M
2
) between the players
ISO and AIS as follows: After stage in the game we have the sequence
f

: ), which is an increasing continuous sequence of partial isomor-

phisms from M
1
to M
2
, and the sequence z

: ) which is an increasing
continuous sequence in T .
Stage in the game is as follows : First, AIS chooses z

of level of T
such that for every < z

>
T
z

. Then:
1. If = 0, then f

= .
2. If is limit, then f

=
<
f

.
3. If = + 1, then AIS chooses A
1
M
1
and A
2
M
2
such that
[A
1
A
2
[ < 1 +. Then ISO should choose f

such that f

is a par-
tial isomorphism fromM
1
to M
2
, f

, A
1
Dom(f

), A
2
Rang(f

)
The rst player who cannot nd a legal move loses the game. If the
isomorphism player ISO has a winning strategy for
T ,
(M
1
, M
2
), we say
that M
1
, M
2
are EF
T ,
equivalent.
LABEL Sec1-Def-
Square-F-lambda)
Denition 2.2. We say that
F,
holds, if
1. T is a set of partial functions from to ;
2. if f is a partial function from to , then f T i fu T for
every countable u Dom(f);
3. there is no f T such that Dom(f) = .
LABEL Sec1-2.3)
Denition 2.3. If
F,
holds, we dene a tree T
F
in the following way:
The nodes are functions f such that f T and Dom(f) is an ordinal;
the order is inclusion.
Note that this tree T
F
does not have a branch of length .
LABEL Sec1-
Remark-Strength-
Of-Game)
Remark 2.4. If T is a tree of size with no -branch, we can assume without
loss of generality that T . Dene T by f T if f is a partial function
from to such that x < y implies f(x) <
T
f(y). We get that
F,
holds,
and T can be embedded (as a partial order) in T
F
.
LABEL Sec1-
Theorem-Eq-T-
lambda)
Theorem 2.5. Suppose
cf() = =

0
,

F,
holds,
T = T
F
.
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


EXISTENCE OF EF-EQUIVALENT NON ISOMORPHIC MODELS 5
Then there are non-isomorphic models M
1
, M
2
of size which are EF
T ,
equivalent.
Proof. First, we shall dene a tool for constructing models.
LABEL Sec2-2.6)
Denition 2.6. x is a structure parameter if it consists of the following
objects:
a set I,
a set J
s
for each s I, such that if s
1
,= s
2
, then J
s
1
J
s
2
=
(denote J =

sI
J
s
),
sets S, T such that S I I and T J J,
LABEL Sec2-2.7)
Denition 2.7. For a given structure parameter x we dene a model M = M
x
in the following way: First for each s I let G
s
be an abelian group gen-
erated freely by x
t
: t J
s
except of the relation x(2x = 0). (We could
have also used a free group or a free abelian group, but this choice makes the
proof a bit simpler). We demand also that if s
1
,= s
2
, then G
s
1
G
s
2
= .
For (s
1
, s
2
) S, let G
s
1
,s
2
be the subgroup of G
s
1
G
s
2
generated by
(x
t
1
, x
t
2
) : (t
1
, t
2
) T (J
s
1
J
s
2
). The universe of M is

sI
G
s
. The
vocabulary of M consists of
1. for each a M, a unary function symbol F
a
;
2. for each s I, a unary relation symbol P
s
;
3. for each (s
1
, s
2
) S, a binary relation symbol Q
s
1
,s
2
.
The interpretation of the symbols in M is as follows:
1. for each b M, s I, a G
s
, if b G
s
, then F
M
a
(b) = a +b, else
F
M
a
(b) = b;
2. for each s I, P
M
s
= G
s
;
3. for each (s
1
, s
2
) S, Q
M
s
1
,s
2
= G
s
1
,s
2
.
LABEL Sec1-
Lemma-Cond-Of-
BeingAuto)
Lemma 2.8. Suppose I

I and f is a function, f :

sI
G
s
M. Then
f is a partial automorphism of M i the following hold:
1. For each s I

, f(0
Gs
) G
s
.
2. For each s I

and a G
s
we have f(a) = f(0
Gs
) +a.
3. For each s
1
, s
2
I

, if (s
1
, s
2
) S, then (f(0
Gs
1
), f(0
Gs
2
)) G
s
1
,s
2
.
Proof. Suppose f is a partial automorphism. Then we have:
1. For each s I

, 0
Gs
G
s
= P
M
s
which implies f(0
Gs
) P
M
s
= G
s
.
2. For each s I

and a G
s
, f(a) = f(F
M
a
(0
Gs
)) = F
M
a
(f(0
Gs
)) =
f(0
Gs
) +a.
3. For each s
1
, s
2
I

, if (s
1
, s
2
) S, then (0
Gs
1
, 0
Gs
2
) G
s
1
,s
2
(be-
cause it is a subgroup of G
s
1
G
s
2
) but G
s
1
,s
2
= Q
M
s
1
,s
2
, therefore
we have (f(0
Gs
1
), f(0
Gs
2
)) G
s
1
,s
2
.
Similar arguments show the other direction.
??
Sec1-
LemmaCondOfBeingAuto
Now we shall dene a structure parameter x and put M = M
x
. Then we will
choose elements a

, b

M, dene M
1
= (M, a

), M
2
= (M, b

), and show
that M
1
, M
2
are as required in Theorem 2.5. Sec1-Theorem-
Eq-T-lambda)
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


6 CHANOCH HAVLIN AND SAHARON SHELAH
Let x = x
,F
be the following structure parameter:
1. I = []

0
.
2. For u I, J
u
consists of the quadruples t = (u, g, h, ), where:
(a) g, h are functions from u into ;
(b) is a function from supRang(g) u into ;
(c) T;
(d) g, h are weakly increasing;
(e) if g(x) = g(y) then h(x) = h(y);
(f) h(x) > x.
For t = (u, g, h, ) we will denote u = u
t
, g = g
t
, h = h
t
, =
t
.
3. S = (u
1
, u
2
) : u
1
, u
2
I and u
1
u
2
.
4. T = (t
1
, t
2
) : t
1
, t
2
J, u
t
1
u
t
2
, g
t
1
g
t
2
, h
t
1
h
t
2
,
t
1

t
2
.
Let M = M
,F
= M
x
be the corresponding model. Note that [I[ =

0
=
and for each u I, [J
u
[ =

0
= , therefore [[M[[ = . Dene a

= 0
G

, b

= x
(,,,)
,
M
1
= (M, a

), M
2
= (M, b

).
LABEL Sec1-
Claim-Equiv-
Models)
Claim 2.9. M
1
, M
2
are EF
T ,
equivalent.
Proof. We start with
LABEL Sec1-Def-
G-of-lambda)
Denition 2.10. We dene a set of functions ( = (() with a partial order

G
in the following way:
1. For an ordinal < , (

is the set of functions g which satisfy

(a) g : , < ;
(b) g is weakly increasing.
2. ( =

<
(

.
3. For each g ( such that Dom(g) = we dene h
g
: + 1 by
h
g
(x) = Min(y : y < g(y) > g(x) ).
4. g
1

G
g
2
if g
1
g
2
and h
g
1
h
g
2
.
LABEL Sec1-
Claim-G-of-
lambda)
Claim 2.11. 1. If g(x) = g(y), h
g
(x) = h
g
(y).
2. h
g
(x) > x.
3. h
g
is weakly increasing.
4. For every g
1
, g
2
(, g
1

G
g
2
i
(a) Dom(g
1
) =
1

2
= Dom(g
2
) and g
1
g
2
;
(b) if
1
<
2
, then g
2
(
1
) > g
2
(x) for every x <
1
.
5. If g
1
(

and Dom(g
1
) < < , then there is g
2
(
+1
such that
g
1

G
g
2
and Dom(g
2
) = .
6. If < and we have g

: < ) such that g

and <
implies g

G
g

, then g =

<
g

satises g (

and g

G
g
for each < .
Proof.
1.-3. Easy.
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


EXISTENCE OF EF-EQUIVALENT NON ISOMORPHIC MODELS 7
4. If there is x <
1
such that g
2
(
1
) = g
2
(x), then h
g
2
(x) = h
g
2
(
1
) >
1
h
g
1
(x),
so g
1

G
g
2
. On the other direction, if g
1
g
2
and g
2
(
1
) > g
2
(x)
for every x <
1
, then for every such x: If there is y <
1
such that
g
1
(y) > g
1
(x), let y

be the minimal y which satises this. We get

h
g
1
(x) = h
g
2
(x) = y

. If there is no such y, we get h

g
1
(x) = h
g
2
(x) =
1
.
Therefore we have h
g
1
h
g
2
.
5. Dene g
2
: + 1 by
g
2
(x) =

g
1
(x) if x Dom(g
1
),
if x Dom(g
1
).
By 4. we get that g
1

G
g
2
.
6. Remember that is regular, therefore

<
Dom(g

) < . This
completes the proof of Claim 2.11.
2.11
Sec1-Claim-G-of-
lambda)
Sec1-Claim-G-of-
lambda)
Now we will describe a winning strategy for ISO in the game
T ,
(M
1
, M
2
).
In stage of the game, ISO will choose a function g

such that
1. g

;
2. if < , then g

G
g

;
3. if is a successor ordinal and in stage AIS chose the sets A
1
, A
2
then for each u I such that (A
1
A
2
)G
u
,= we have u Dom(g

).
The choice of g

is done in the following way:

1. g
0
= .
2. If is limit, then g

=
<
g

. By Claim 2.11, g

and if Sec1-Claim-G-of-
lambda) < then g

G
g

.
3. If = +1 and in stage AIS chose the sets A
1
, A
2
, ISO will choose
< such that Dom(g

) < and u for every u I such that

(A
1
A
2
) u ,= (such exists sence [A
1
A
2
[ +
0
< ) . By
Claim 2.11 there is g (

such that Dom(g) = and g

G
g. ISO Sec1-Claim-G-of-
lambda) will choose such a function as g

.
Now remember that if = +1, then in stage AIS has to choose a node on
level , which is actually a function

: ,

T. Then he chooses
A
1
M
1
and A
2
M
2
. Then ISO has to choose a partial isomorphism
f

from M
1
to M
2
such that f

, A
1
Dom(f

), A
2
Rang(f

)
(see Denition 2.1). So, ISO chooses g

, and then denes f

according to Sec1-Def-Game-
With-Tree) f

, A
1
, A
2
, g

in the following way :

Dom(f

) = Dom(f

G
u
: u I, (A
1
A
2
) G
u
,= .
Then, for each u I we have G
u
Dom(f

) or G
u
Dom(f

) = .
If G
u
Dom(f

), we dene f

(0
Gu
) = x
t
, where t = (u, g

u, h
g
u,

(u supRang(g

u))).
(Note that because g

, we have Rang(g

) = Dom(

).)
Next, for every a G
u
we dene f

(a) = f

(0
Gu
) +a. By the construc-
tion we get that if (u
1
, u
2
) S, then (f

(0
Gu
1
), f

(0
Gu
2
)) G
u
1
,u
2
(because
the corresponding couple of ts lies in T). Therefore by Lemma ??, f

is a Sec1-
LemmaCondOfBeingAuto partial automorphism of M. We also have:
1. If < , then g

, h
g

h
g
and

. Therefore f

.
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


8 CHANOCH HAVLIN AND SAHARON SHELAH
2. For each > 0,
f

(a

) = f

(0
G

) = x
(,,,)
= b

.
Therefore f

is a partial isomorphism from M

1
= (M, a

) into M
2
=
(M, b

).
This completes the proof of Claim 2.9.
2.9
Sec1-Claim-
Equiv-Models)
Sec1-Claim-
Equiv-Models)
LABEL Sec1-
ClaimNotIso)
Claim 2.12. M
1
, M
2
are not isomorphic.
Proof. It is enough to show that M is rigid (i.e. it does not have a non-trivial
automorphism).
Assume towards contradiction that f ,= id is an automorphism of M. For
each u I we dene c
u
= f(0
Gu
). By Lemma 2.8, for each u w I we
Sec1-Lemma-
Cond-Of-
BeingAuto)
have (c
u
, c
w
) G
u,w
.
For each u w I and t = (w, g, h, ) J
w
we dene
w,u
(t) J
u
by

w,u
(t) =: (u, gu, hu, supRang(gu) u).
By the denition of T we have that if t J
w
and r J
u
, then (r, t) T i
r =
w,u
(t). We dene a homomorphism
w,u
: G
w
G
u
by
w,u
(x
t
) = x
r
,
where r =
w,u
(t). We get that G
u,w
is the subgroup of G
u
G
w
generated
by (
w,u
(x
t
), x
t
) : t J
w
. Since x
t
: t J
w
generate G
w
, we get that
G
u,w
= (
w,u
(c), c) : c G
w
.
Now dene n(u) to be the length of the reduced representation of c
u
as a
sum of the generators x
t
: t J
u
. For u w I we get n(u) n(w),
since c
u
=
w,u
(c
w
) and
w,u
sends one generator to one generator. If for
every u I there is w I such that n(w) > n(u), we can nd a sequence
u
n
: n < ) such that u
n
I and n(u
n
) < n(u
n+1
). Dene w =

n<
u
n
,
we get that n(w) is innite - contradiction. Therefore, there is u

I such
that n(u

) is maximal. Since we assumed f ,= id , n(u

) > 0.
Choose t

J
u
such that x
t
appears in the reduced representation of c
u
.
For each u

w I there is a unique t(w) J

w
such that
w,u
(t(w)) = t

and x
t(w)
appears in the reduced representation of c
w
. Such t(w) exists
because c
u
=
w,u
(c
w
). It is unique because if there were two such ts,
t
1
, t
2
, then

w,u
(x
t
1
) =
w,u
(x
t
2
) = x
t
.
Since in G
u
x(2x = 0), it implies n(w) > n(u

), which contradicts the

maximality of n(u

).
Note that if u w z I, then
z,u
=
w,u

z,w
. Therefore, by unique-
ness of t(w), if u

w z I then t(w) =
z,w
(t(z)). For each u

w I,
dene g
w
= g
t(w)
, h
w
= h
t(w)
,
w
=
t(w)
. If u

w
1
, w
2
I, then the
functions g
w
1
, h
w
1
,
w
1
and g
w
2
, h
w
2
,
w
2
are respectively compatible, since
t(w
1
) =
z,w
1
(t(z)) and t(w
2
) =
z,w
2
(t(z)),
where z = w
1
w
2
. Dene
g = g
w
: u

w I, h = h
w
: u

w I, =
w
: u

w I.
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


EXISTENCE OF EF-EQUIVALENT NON ISOMORPHIC MODELS 9
We get:
1. Dom(g) = Dom(h) = .
2. g, h are weakly increasing.
3. h(x) > x.
4. If g(x) = g(y), then h(x) = h(y).
5. T (this is by Denition 2.2(2)). Sec1-Def-Square-
F-lambda) 6. supRang(g) Dom().
By Denition 2.2(3), Dom() ,= . Therefore by 6., supRang(g) < . Since Sec1-Def-Square-
F-lambda) g is weakly increasing and is regular, there is
0
< such that for every

0
< < , g() = g(
0
). By 4. we get that for every
0
< < , h() = h(
0
).
Choose > h(
0
) >
0
and get that h() < , contradicting 3.
This completes the proof of Claim 2.12.
2.12
The proof of Theorem Sec1-
ClaimNotIso)
Sec1-
ClaimNotIso)
2.5 is now nished.
2.5
Sec1-Theorem-
Eq-T-lambda)
Sec1-Theorem-
Eq-T-lambda)
3. Games with trees for singular =

0
It is clear that for singular we cannot expect the same result as in the
previous section, since the AIS player would be able to list all the members
of M
1
, M
2
. Thus, we prove a weaker result - we allow AIS to choose only
one element in each turn. We also remark in Remark 3.2 that this result
Section2-
Generalize-866)
generalizes the result in [4] for such .
LABEL Sec2-
Theorem-Eq-T,1)
Theorem 3.1. Suppose
cf() < =

0
,

F,
holds,
T = T
F
.
Then there are non-isomorphic models M
1
, M
2
of size which are EF
T ,1
equivalent.
LABEL Section2-
Generalize-866)
Remark 3.2. We can show that Theorem 3.1 generalizes the result in [4] by
Sec2-Theorem-
Eq-T,1)
choosing appropriate T. The result there shows the existence of two non-
isomorphic models of size which are equivalent under every EF game of
length < cf(), which consists of sub-games of length < , such that AIS
chooses the length of each sub-game before it starts, and in every sub-game
he chooses one element in each move - see the denitions there. Now, an
appropriate T can be chosen by looking at the proof there, but we will take
a shortcut - we will use the result instead of the proof. Let us choose a pair
of models M
1
, M
2
as in the result in [4]. Without loss of generality assume
that the universe of M
1
is 1 and the universe of M
2
is 2. We can
take T to be the set of functions f which satisfy the following conditions:
1. Dom(f) , Rang(f) .
2. The partial function f

from M
1
to M
2
dened by
Dom(f

) = Dom(f)1, where for every Dom(f), f

((, 1)) = (f(), 2)

is a partial isomorphism.
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


10 CHANOCH HAVLIN AND SAHARON SHELAH
Now, it is not hard to see that EF
T
F
,1
equivalence implies equivalence as in
the result of [4].
Proof of theorem 3.1. Denote = cf(). ( >
0
because =

0
.) Let Sec2-Theorem-
Eq-T,1)
i
: i < ) be an increasing and continuous sequence such that
0
= 0,

+
i
<
i+1
= cf(
i+1
), g
i
>
0
for i > 0, and

i<

i
= . For every
< there is a unique i < such that [
i
,
i+1
). We denote i = i().
We dene a structure parameter x = x
F,
in the following way:
1. I = []

0
.
2. For u I, J
u
is the collection of quadruples t = (u, g, h, ) such that
(a) g, h are functions from u into , is a function from some subset
of u into ;
(b) T;
(c) for every x u, g(x) [
i(x)
,
+
i(x)
], h(x) [
i(x
),
i(x)+1
];
(d) g, h are weakly increasing;
(e) if g(x) = g(y), then h(x) = h(y);
(f) h(x) > x;
(g) Dom() = u

i(x)
: x u and h(x) =
i(x)+1

For t = (u, g, h, ) we denote u = u

t
, g = g
t
, h = h
t
, =
t
.
3. S = (u
1
, u
2
) : u
1
, u
2
I, u
1
u
2
.
4. T = (t
1
, t
2
) J : u
t
1
u
t
2
, g
t
1
g
t
2
, h
t
1
h
t
2
,
t
1

t
2
.
Let M = M
F,
= M
x
be the corresponding model. Dene a

= 0
G

, b

= x
(,,,)
.
Dene M
1
= (M, a

) and M
2
= (M, b

).
LABEL Sec2-
Claim-Equiv-
Models)
Claim 3.3. M
1
, M
2
are EF
T ,1
equivalent.
Proof. We start with
LABEL Sec2-Def-
W)
Denition 3.4. A partially ordered set of functions (J,
W
), which de-
pends on the sequence
i
: i < ), is dened in the following way:
1. we dene a set B such that

B i
(a)

=
i
: i < ),
i

i

i+1
;
(b) there is j = j(

) < such that i < j(

) i
i
=
i+1
.
2. For

B we dene J

to be the set of functions g which satisfy

(a) Dom(g) =
i<
[
i
,
i
);
(b) g is weakly increasing;
(c) for every i < , x [
i
,
i
), we have g(x) [
i
,
+
i
], and if
g(x) =
+
i
, then i < j(

).
3. For j < we dene J
j
=

: j() j.
4. For g J

we dene a function h
g
as follows: Dom(h
g
) = Dom(g)
where for i < and x [
i
,
i
) we have h
g
(x) = Min(y :
i
y <
i
g(y) > g(x)
i
).
LABEL Sec2-
Claim-W)
Claim 3.5. 1. If g(x) = g(y), then h
g
(x) = h
g
(y).
2. h
g
(x) > x.
3. h
g
is weakly increasing.
4. If x [
i
,
i+1
), then h
g
(x) [
i
,
i+1
].
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


EXISTENCE OF EF-EQUIVALENT NON ISOMORPHIC MODELS 11
5. Suppose that g
1
J

1 , g
2
J

2. Then g
1

W
g
2
i
(a) g
1
g
2
(therefore for every i <
1
i

2
i
);
(b) for every i < if
1
i
<
2
i
, then for every x [
i
,
1
i
), g
2
(x) <
g
2
(
1
i
).
6. If g
1
J
j
and

B, j(

) j, then there is g
2
J
j
such that
g
1

W
g
2
and

i<
[
i
,
i
) Dom(g
2
).
7. If <
+
j
and g

: < ) is such that g

J
j
and < implies
g

W
g

then there exists g J

j
such that if < , then g

W
g.
Proof.
1. - 4. Easy.
5. Like in the proof of Claim 2.11. Sec1-Claim-G-of-
lambda) 6. We may assume that Dom(g
1
)

i<
[
i
,
i
). Dene for i <

i
=
i
+ Supg
1
(x) : x Dom(g
1
) [
i
,
i+1
).
Since g
1
J
j
we have
i
<
+
i
for i j. Dene for i < ,

i
=

+
i
if i < j,

i
+ 1 if i j
.
Now dene g
2
with Dom(g
2
) =

i<
[
i
,
i
), where for every i <
and x [
i
,
i
),
g
2
(x) =

g
1
(x) if x Dom(g
1
),

i
if x / Dom(g
1
)
Since j(

) j we have g
2
J
j
. By 5. we have g
1

W
g
2
.
7. Dene for every i < ,

i
= sup(

<
Dom(g

)[
i
,
i+1
) )+
i
,
i
= sup(

<
Rang(g

[
i
,
i+1
))+
i
,
For every < , g

J
j
. Therefore for every i j,
sup(Dom(g

) [
i
,
i+1
)) <
i+1
, supRang(g

[
i
,
i+1
)) <
+
i
.
Therefore, since <
+
j

+
i
<
i+1
= cf(
i+1
), we get that for i j
we have
i
<
i+1
and
i
<
+
i
.
Dene for i < ,

i
=

i+1
if i < j,

i
if i j,

i
=

+
i
if i < j,

i
+ 1 if i j.
Denote g

<
g

. Dene g J
j
with Dom(g) =

i<
[
i
,

i
)
for i < and x [
i
,

i
) by
g(x) =

(x) if x Dom(g

),

i
if x / Dom(g

).
By 5. we get that < implies g
W
g

.
This completes the proof of Claim 3.5.
3.5
Sec2-Claim-W)
Sec2-Claim-W)
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


12 CHANOCH HAVLIN AND SAHARON SHELAH
Now we will describe a winning strategy for ISO:
In every stage in the game ISO will choose a function g

such that
1. g

J
i()+1
;
2. if < , then g

W
g

;
3. if in stage AIS chose an element from G
u
, then u Dom(g

).
ISO can choose such g

in the following way:

1. For = 0, g
0
= .
2. For limit, since <
i()+1
and for every < , g

J
i()+1
, we
can use Claim 3.5(7). Sec2-Claim-W)
3. If = + 1 and in stage AIS chose an element from G
u
, then we
choose

=
i
: i < ) in the following way : If i < i() + 1, then

i
=
i+1
. Else,
i+1
> . We choose
i
<
i+1
such that
u [
i
,
i+1
) [
i
,
i
).
Now j(

) = i() + 1, so by Claim 3.5(6) we can nd g J

i()+1
Sec2-Claim-W)
such that
g

W
g and

i<
[
i
,
i
) Dom(g).
Dene g

= g.
Now if = + 1 and in stage AIS chose an element from G
u
and the
node

T , ISO will dene the automorphism f

according to g

with Dom(f

) = Dom(f

) G
u
. For every w such that G
w
Dom(f

),
f

(0
Gw
) = x
t
where
t = (w, g

w, h
g
w,

(w
i(x)
: x w h
g
(x) =
i(x)+1
))
(Note that v = Dom(

), because g

J
i()+1
.) As in Section 2,
we get that f

is a partial isomorphism and < implies f

. This
completes the proof of Claim 3.3.
3.3
Sec2-Claim-
Equiv-Models)
Sec2-Claim-
Equiv-Models)
LABEL Sec2-
ClaimNotIso)
Claim 3.6. M
1
, M
2
are not isomorphic.
Proof. We imitate the proof of Claim 2.12. It is enough to show that M
Sec1-
ClaimNotIso)
is rigid. Assume towards contradiction that f ,= id is an automorphism of
M. For each u w I and t = (w, g, h, ) J
w
we dene
w,u
(t) J
u
by

w,u
(t) = (u, g
t
u, h
t
u,
t
v), where v =

i(x)
: x u h
t
(x) =
i(x)+1
u.
We proceed as in the proof of Claim 2.12, and we get that we can nd
Sec1-
ClaimNotIso)
functions g, h, such that the following hold:
1. Dom(g) = Dom(h) = , Dom() .
2. If i(x) = i, then g(x) [
i
,
+
i
], h(x) [
i
,
i+1
].
3. g, h are weakly increasing.
4. If g(x) = g(y), then h(x) = h(y).
5. h(x) > x.
6. If h(x) =
i(x)+1
, then
i(x)
Dom().
7. T.
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


EXISTENCE OF EF-EQUIVALENT NON ISOMORPHIC MODELS 13
By 7. we get that Dom() ,= , therefore by 6. there exists i < such that
if i(x) = i, then i(h(x)) = i. By 2., i(x) = i implies g(x)
+
i
. By 3., g is
weakly increasing. Since
i+1
= cf(
i+1
) >
+
i
, we can nd
0
such that if

0
x <
i+1
, then g(x) = g(
0
). By 5., h(
0
) >
0
. By the choice of i we
get that h(
0
) <
i+1
. Choose h(
0
) < x <
i+1
. We get h(x) > x > h(
0
)
but g(x) = g(
0
). This contradicts 4. Therefore we proved that M is rigid.
This completes the proof of Claim 3.6.
3.6
Sec2-
ClaimNotIso)
Sec2-
ClaimNotIso)
Ths proof of Theorem 3.1 is now also completed.
3.1
Sec2-Theorem-
Eq-T,1)
Sec2-Theorem-
Eq-T,1)
4. regular >

In this section we show a result which holds for every regular >

. In
the previous sections we used the assumption =

0
. Here we use instead
of it the existence of a set T []

0
of size which is dense. By dense
we mean that for every A []

there is B A, B T.
LABEL Sec4-4.1)
Remark 4.1. 1. Looking at the proof, one can see that instead of >

,
it is enough to assume the following:
(a) > 2

0
.
(b) There is T []

0
such that
i. [T[ = ;
ii. for every A []

, there is B T, such that B A.

2. It is possible that it can be proved in ZFC that every > 2

0
satises
1.(b) (it is a problem in cardinal arithmetic).
LABEL Sec3-
Theorem-Equiv)
Theorem 4.2. Suppose
= cf() >

,
T is a tree of size with no branch of length .
Then there are models M
1
, M
2
of size which are EF
T ,
equivalent but not
isomorphic.
Proof. Let be large enough cardinal (for example =
7
()).
LABEL Sec3-
Claim-exists-
inner-model)
Claim 4.3. We can nd M such that the following hold:
1. M is elementary sub-model of H().
2. + 1 M.
3. [[M[[ = .
4. For every (x
i
, z
i
) : i < ) such that x
i
M and z
i
T for every
i < there exists an increasing sequence i
n
: n < ) such that
(a) (x
in
, z
in
) : n < ) M;
(b) if in addition for i < j < the level of z
i
(in T ) is strictly less
than the level of z
j
, then z
in
: n < ) is an antichain in the
order
T
.
In the proof Claim 4.3 we use a partial version of the RGCH Theorem Sec3-Claim-
exists-inner-
model)
(see Shelah [5]).
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


14 CHANOCH HAVLIN AND SAHARON SHELAH
LABEL RGCH-
Theorem)
Theorem 4.4. (RGCH Theorem, partial version). If

, then there is
regular <

and T []
<
such that
1. [T[ = ,
2. for every A []

, we can nd A
i
: i < ) such that < , A
i
T
for every i < , and A =

i<
A
i
.
LABEL RGCH-
Corollary)
Corollary 4.5. If

, then we can nd a set T

[]

0
such that
[T

[ = and for every A []

there is B T

such that B A.
Proof. Choose and T as in Theorem 4.4 and dene T

[A]

0
: A RGCH-Theorem)
T.
4.5
RGCH-Corollary)
Proof of Claim 4.3. We construct M
n
for every n < such that
Sec3-Claim-
exists-inner-
model)
1. M
0
is an elementary sub-model of H() such that [[M
0
[[ = , +1
M
0
, and for every A []

there is B M
0
[]

0
such that B A
(this is possible by Corollary 4.5);
RGCH-Corollary)
2. [[M
n
[[ = ;
3. M
n
is an elementary sub-model of H();
4. if A M
n
and [A[ , then A M
n+1
;
5. M
n
M
n+1
, and M
n
M
n+1
.
Now, let M =

n<
M
n
. We will prove that M satises the conclusion
of Claim 4.3. Sec3-Claim-
exists-inner-
model)
Suppose that (x
i
, z
i
) : i < ) MT for every i < . We may assume
without loss of generality that there is n
0
< such that (i, x
i
, z
i
) : i <
M
n
0
. If the condition in Claim 4.34.(b) is not satised, then we are
Sec3-Claim-
exists-inner-
model)
done, because we can nd A []

0
such that (i, x
i
, z
i
) : i A M
n
0
+1
.
(because in M
n
0
+1
there is a one to one correspondence between M
n
0
T
and , and every subset of of size

has innite countable subset that is

a member of M
0
).
If the condition in Claim 4.34.(b) is satised, then we have two cases: Sec3-Claim-
exists-inner-
model)
Case (1): We can nd A []

such that z
i
: i A) is an antichain in

T
.
Case (2): We cannot nd such A.
If we are in Case(1), then we are done in the same way as before.
Suppose we are in Case(2).
LABEL Sec3-
small-1)
Claim 4.6. For every j < , we can nd j < i
0
< i
1
< i
2
< , such that
z
i
0
<
T
z
i
1
, z
i
2
and z
i
1
, z
i
2
are not comparable in
T
.
Proof. Assume towards contradiction that there is j < such that we
cannot nd j < i
0
< i
1
< i
2
< which are as in the claim. Dene
C = z
i
: j < i < . Then comparability in
T
is an equivalence relation
on C. Since is regular, either there are equivalence classes or there is
an equivalence class of size . In other words, C contains an antichain or a
chain of size . Both options are not possible, the rst since we are in Case
(2) and the second since T does not have a branch. Contradiction.
4.6
Sec3-small-1)
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


EXISTENCE OF EF-EQUIVALENT NON ISOMORPHIC MODELS 15
By Claim 4.6 we can choose for every j < a triple i
0
(j), i
1
(j), i
2
(j) such Sec3-small-1)
that
1. i
0
(j) < i
1
(j) < i
2
(j) < ;
2. j < j

i
2
(j) < i
0
(j

);
3. z
i
0
(j)
<
T
z
i
1
(j)
, z
i
2
(j)
;
4. z
i
1
(j)
and z
i
1
(j)
are not comparable in
T
.
We choose A []

0
such that (j, i
0
(j), i
1
(j), i
2
(j), x
j
, z
j
) : j A M
n
0
+1
.
Using the Ramesy Theorem in M
n
0
+1
, we can nd an increasing sequence
j
n
: n < ) such that
1. j
n
A for every n < ;
2. j
n
: n < ) M
n
0
+1
;
3. z
i
1
(jn)
: n < is a chain or an antichain in T ;
4. z
i
2
(jn)
: n < is a chain or an antichain in T .
Now we are done, since either z
i
1
(jn)
: n < or z
i
1
(jn)
: n < must be
an antichain. Because if both are chains, we get that z
i
1
(j
0
)
<
T
z
i
1
(j
1
)
, z
i
2
(j
0
)
<
T
z
i
2
(j
1
)
.
Since z
i
0
(j
1
)
is on higher level then z
i
1
(j
0
)
, z
i
2
(j
0
)
and it is <
T
z
i
1
(j
1
)
, z
i
2
(j
1
)
we get that z
i
1
(j
0
)
, z
i
2
(j
0
)
<
T
z
i
0
(j
1
)
- contradiction, since by the construction
they are not comparable.
This completes the proof of Claim 4.3.
4.3
Sec3-Claim-
exists-inner-
model)
Sec3-Claim-
exists-inner-
model)
We choose M as in Claim 4.3 and we dene a structure parameter x = x(M)
Sec3-Claim-
exists-inner-
model)
in the following way:
LABEL Sec3-Def-
Parameter)
Denition 4.7. 1. I consists of the objects of the form (u, ), where
(a) u
<
0
;
(b) M, [[
0
, is a set of partial functions from to
with nite domain.
For s = (u, ) we denote u = u
s
and =
s
. We dene (s) =
u
s

Dom(f) : f
s
. Note that this a countable set.
2. For s = (u, ) I, J
s
consists of all the objects of the formt = (u, , g, h, F, z),
where
(a) g, h are functions from u to ;
(b) F is a function from
2
to 0, 1;
(c) z T ;
(d) the level of z in the tree T is minimal under the condition
> y for every y such that y Rang(g) or there are f
1
, f
2

such that F(f
1
, f
2
) = 1 and y Rang(f
1
):
(e) there is a witness (g, h) for t, which means that
i. Dom(g) = Dom(h) , Rang(g) Rang(h) ,
ii. (s) Dom(g).
iii. g, h are weakly increasing,
iv. h(x) > x,
v. if g(x) = g(y), then h(x) = h(y),
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


16 CHANOCH HAVLIN AND SAHARON SHELAH
vi. g g, h h,
vii. for every (f
1
, f
2
)
2
, F(f
1
, f
2
) = 1 i f
1
g and
f
2
h.
3. S = I
2
.
4. T consists of the pairs (t
1
, t
2
) J
2
, where
(a) t
1
, t
2
have a common witness;
(b) z
t
1
, z
t
2
are comparable in the order
T
.
LABEL Sec4-4.8)
Fact 4.8. s I, z T ,
g, h satisfy conditions i.-v. from Denition 4.72.(e), Sec3-Def-
Parameter) Dom(g) , where is the level of z.
Then the following hold:
1. There is unique t J
s
such that (g, h) is a witness for t and z
t

T
z.
We denote t = t(s, g, h, z).
2. If
(a) g

, h

, z

also satisfy the conditions in 1.,

(b) z, z

are comparable in
T
,
(c) g

, h

are compatible with g, h respectively.

then t(s, g, h, z) = t(s, g

, h

, z

).
Let M = M
x
be the corresponding model. We can check that [[M[[ = .
Let a

= 0
G
(,)
, b

= x
(,,,,,z)
where, z

is the root of T (without loss

of generality there is a root). Dene M
1
= (M, a

), M
2
= (M, b

).
LABEL Sec3-
Claim-Equiv)
Claim 4.9. M
1
, M
2
are EF
T ,
equivalent.
We describe a winning strategy for ISO - this is very similar to the proof
of Claim 2.9, so we will omit the details. We are using the denitions in Sec1-Claim-
Equiv-Models) Denition 2.10.
Sec1-Def-G-of-
lambda)
In every stage of the game ISO will choose a function g

such that the

following hold:
1. g
0
= .
2. g

(see denition of (

and
G
in Denition 2.10). Sec1-Def-G-of-
lambda) 3. < implies g

G
g

.
4. If in stage AIS chose the sets A
1
, A
2
, then for each s I, if
G
s
(A
1
A
2
) ,= , then (s) Dom(g

).
Now if = +1 and in stage AIS chose the sets A
1
, A
2
and the node z

,
ISO will dene h

= h
g
and then dene f

by
1. Dom(f

) =

G
s
: (s) Dom(g

),
2. for each s such that G
s
Dom(f

), f

(0
Gs
) = x
t
, where t =
t(s, g

, h

, z

).
LABEL Sec3-
Claim-Not-Iso)
Claim 4.10. M
1
, M
2
are not isomorphic.
Proof. It is enough to show that M is rigid. Assume towards contradiction
that f ,= id is an automorphism of M. Denote c
s
= f(0
Gs
) for s I.
Denote W
s
= t J
s
: x
t
is in the reduced representation of c
s
. Since
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


EXISTENCE OF EF-EQUIVALENT NON ISOMORPHIC MODELS 17
f ,= id there is s

= (u

) such that W
s
,= . Note also that if u
s

u
s
and

s
, then there is a natural projection
s,s
from J
s
into J
s
such
that W
s
Rang(
s,s
W
s
) (see the proof of Claim 2.12), therefore W
s
,= . Sec1-
ClaimNotIso) Choose s
i
, t
i
,
i
for i < such that the following hold:
1. s
i
I, s
i
= (u

i
,

)
2. t
i
W
s
i
3.
i
<
4. If i < j h
t
i
(
i
) <
j
.
Case (*1) : supg
t
i
(
i
) : i < = . Then, since the level of z
t
i
in T must
be greater then g
t
i
(
i
), we may assume that if i < j, then the level of z
t
i
is
strictly less than the level of z
t
j
.
Case (*2): supg
t
i
(
i
) : i < < . Then by regularity of , we may
assume that for every i, j < , g
t
i
(
i
) = g
t
j
(
j
).
Now, no matter in which case we are, we proceed in the following way:
By the properties of M (see Claim 4.3) we can nd a set A such that Sec3-Claim-
exists-inner-
model)
1. [A[ =
0
;
2. W
s
i
: i A M;
3. if we are in Case (*1), z
t
i
: i A is an antichain (we can have
that because in Case(*1) the level of z
t
i
is strictly increasing with i
- see Claim 4.3). Sec3-Claim-
exists-inner-
model)
We dene s
+
= (u

, g
t
, h
t
: t W
s
i
, i A

). (Note that

iA
W
s
i

M, therefore s
+
I.)
LABEL Sec3-
Claim-unique-
projection)
Claim 4.11. For every i A, if r J
s
+, t W
s
i
, (r, t) T, then we have:
1. If (g, h) is a witness for r, then g
t
g, h
t
h.
2. If t ,= t

J
s
i
, then (r, t

) / T.
Proof.
1. Let (g
0
, h
0
) be a common witness for r, t. Then we have g
t

g
0
, h
t
h
0
. Now g
t
, h
t

s
+
, therefore (g
t
, h
t
) Dom(F
r
). Since
(g
0
, h
0
) is a witness for r and g
t
g
0
, h
t
h
0
, then F
r
(g
t
, h
t
) = 1.
Therefore for any witness (g, h) of r, we have g
t
g, h
t
h.
2. There are three cases:
(a) g
t
,= g
t

or h
t
,= h
t

. Then, since all those functions have the

same domain, we get that r, t

cannot have a common witness

(g, h) because by 1. we must have g
t
g, h
t
h.
(b) F
t
,= F
t

. Then, since Dom(F

t
) =

s
+
= Dom(F
r
) and
(r, t) T we know that F
t
F
r
. Since F
t
,= F
t

and Dom(F
t
) =
Dom(F
t

), we get that F
r
and F
t

are not compatible (and

therefore there is no common witness).
(c) z
t
,= z
t

. By the previous cases we may assume that F

t
=
F
t

, g
t
= g
t

, and h
t
= h
t

, therefore z
t
, z
t

are on the same level

(see Denition 4.7,2.(d)). We can also see that z
r
must be on Sec3-Def-
Parameter) a greater level (remember that F
t
F
r
and F
r
(g
t
, h
t
) = 1).
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


18 CHANOCH HAVLIN AND SAHARON SHELAH
Since (r, t) T, z
t
, z
r
are comparable in
T
. It follows that
z
t

, z
r
are not comparable, thus (r, t

) / T.

4.11
Sec3-Claim-
unique-projection)
LABEL Sec3-
Claim-exists-
projection)
Claim 4.12. For every i A there is r W
s
+ such that (r, t
i
) T.
Proof. Since (c
s
, c
s
+) G
s,s
+ and this group is generated by (x
t
, x
t
) :
(t, t

) T (J
s
J
s
+), there are representations(not necessarily reduced)
c
s
i
= x
w
1
+ +x
wn
, c
s
+ = x
r
1
+ +x
rn
with (r
n
, w
n
) T.
We may assume that if 1
1
<
2
n, then either r

1
,= r

2
or w

1
,= w

2
.
(Otherwise, we can reduce both representations - remember that in those
groups 2x = 0). Since x
t
i
appears in the reduced representation of c
s
i
, t
i
must appear among the ws. Let be such that w

= t
i
. Now we show that
if
1
,= , then r

1
,= r

. Assume toward contradiction that r

1
= r

. By our
assumption, w

1
,= w

. Now, we have:
1. (r

1
, w

1
), (r

, w

) T;
2. w

W
s
i
;
3. w

,= w

1
.
This contradicts Claim 4.11. Sec3-Claim-
unique-projection) We got that for every
1
,= , r

1
,= r

, which implies that x

r

does not
cancel. Hence r

W
s
+ and we are done.
4.12
Sec3-Claim-
exists-projection)
Now choose r
i
W
s
+ for each i A such that (r
i
, t
i
) T.
LABEL Sec3-
claim-r-ies-
distinct)
Claim 4.13. If i < j, then r
i
,= r
j
.
Proof. If we are in Case (*1), then z
t
i
: i A is an antichain. So, z
t
i
, z
t
j
are not comparable. Since z
r
i

T
z
t
i
and z
r
j

T
z
t
j
(see the proof of
Claim 4.11 - z
r
i
, z
t
i
are comparable and z
r
i
is on greater level), we must Sec3-Claim-
unique-projection) have r
i
,= r
j
.
If we are in Case (*2), assume toward contradiction that r = r
i
= r
j
. Let
(g, h) be a witness for r. Then by Claim 4.11, g
t
i
, g
t
j
g, h
t
i
, h
t
j
h. Sec3-Claim-
unique-projection) Since we are in Case (*2) we get that g(
i
) = g(
j
) but by the construc-
tion h(
i
) <
j
< h(
j
), which contradicts the denition of a witness (see
Denition 4.7,2.(e)).
4.13
Sec3-Def-
Parameter)
Sec3-claim-r-ies-
distinct)
We got that W
s
+ is innite - contradiction. Therefore M must be rigid, and
hence the proof of Claim 4.10 is nished..
4.10
Sec3-Claim-Not-
Iso)
Sec3-Claim-Not-
Iso)
With the proof of Claim 4.10 the proof of Theorem 4.2 is also completed.
Sec3-Claim-Not-
Iso)
Sec3-Theorem-
Equiv)

4.2
Sec3-Theorem-
Equiv)
5. > cf() >

Clearly, for being singular and >

we cannot prove the same result

as for regular >

(since in such game AIS will be able to list all the

elements of the two models). Therefore, we dene another type of game.
LABEL Sec5-5.1)
Denition 5.1. Let M
1
, M
2
be models with common vocabulary. Let T be
a tree. The game

T
(M
1
, M
2
) is dened in the same way as the denition
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


EXISTENCE OF EF-EQUIVALENT NON ISOMORPHIC MODELS 19
of
T ,
(see 2.1) except that in stage we demand that the sets A
1
, A
2
Sec1-Def-Game-
With-Tree)
chosen by AIS will satisfy [A
1
A
2
[ < 1 + instead of [A
1
A
2
[ < 1 + .
We say that M
1
, M
2
are EF

T
equivalent if ISO has a winning strategy for
EF

T
(M
1
, M
2
).
LABEL Sec5-5.2)
Remark 5.2. Note that in Theorem 3.1, if we replace EF
T ,1
with EF

T
Sec2-Theorem-
Eq-T,1)
we do not get a stronger result, because for every tree T which satises
the conditions there, we can construct another tree T

which satises the

conditions, so that EF
T

,1
equivalence would imply EF

T
equivalence.
LABEL Sec4-
Theorem-EF*-T-
Equiv)
Theorem 5.3. Suppose that:
1. > cf() = >

;
2. T is a tree of size without a branch.
Then there are non-isomorphic models M
1
, M
2
of size which are EF

T
equivalent.
Proof. Let be a large enough cardinal(for example =
7
()).
LABEL Sec4-
Claim-exists-
inner-model)
Claim 5.4. We can nd M such that the following hold:
1. M is elementary sub-model of H().
2. + 1 M.
3. For every (x
i
, z
i
) : i < ) such that x
i
M and z
i
T for every
i < there exists an increasing sequence i
n
: n < ) such that :
(a) (x
in
, z
in
) : n < ) M;
(b) if in addition for every < there is i < such that the level
of z
i
is greater than , then we can also have that z
in
: n < )
is an antichain in
T
.
Proof. The same proof as the proof of Claim 4.3 (we are using the fact that Sec3-Claim-
exists-inner-
model)
is regular and >

).
5.4
Sec4-Claim-
exists-inner-
model)
Let M be as in Claim 5.4. Let
i
: i < ) be an increasing and continuous
Sec4-Claim-
exists-inner-
model)
sequence such that
0
= 0,
+
i
+
0
<
i+1
= cf(
i+1
), and

i<

i
= .
For every < there is a unique i < such that [
i
,
i+1
). We
denote this i by i().
We dene a structure parameter x in the following way:
LABEL Sec4-Def-
Parameter)
Denition 5.5. 1. I consists of the objects of the form (u, ), where:
(a) u
<
0
;
(b) M, [[
0
, is a set of partial functions from to
with nite domain.
For s = (u, ) we denote u = u
s
, =
s
. We dene (s) = u
s

Dom(f) : f
s
. Note that this is a countable set.
2. For s = (u, ) I, J
s
consists of the objects of the formt = (u, , g, h, F, z),
where
(a) g, h are functions from u to ;
(b) F is a function from
2
to 0, 1;
(c) z T ;
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


20 CHANOCH HAVLIN AND SAHARON SHELAH
(d) the level of z in the tree T . then is minimal with regard
to the condition that
i(x)
for every x such that h(x) =

i(x)+1
or there are f
1
, f
2
such that F(f
1
, f
2
) = 1 and
f
2
(x) =
i(x)+1
;
(e) there is a witness (g, h) for t, which means that
i. Dom(g) = Dom(h) , Rang(g) Rang(h) ,
ii. (s) Dom(g),
iii. g g, h h,
iv. for every (f
1
, f
2
)
2
F(f
1
, f
2
) = 1 i f
1
g f
2
h,
v. g, h are weakly increasing,
vi. h(x) > x,
(i) if g(x) = g(y), then h(x) = h(y),
viii. g(x) [
i(x)
,
+
i(x)
],
ix. h(x) [
i(x)
,
i(x)+1
].
3. S = I
2
.
4. T consists of the pairs (t
1
, t
2
) J
2
, where :
(a) t
1
, t
2
have a common witness;
(b) z
t
1
, z
t
2
are comparable in the order
T
.
Fact 5.6. Suppose
s I, z T ,
g, h satisfy i-ii, v-ix from Denition 5.52.(e), Sec4-Def-
Parameter)

i(x)
: h(x) =
i(x)+1
, where is the level of z.
Then the following hold:
1. There is a unique t J
s
such that (g, h) is a witness for t and
z
t

T
z. We denote t = t(s, g, h, z).
2. If
(a) g

, h

, z

satisfy the conditions in 1.,

(b) z, z

are comparable in
T
,
(a) g

, h

are compatible with g, h, respectively,

then t(s, g, h, z) = t(s, g

, h

, z

).
Let M = M
x
be the corresponding model. We can check that [[M[[ = .
Let a

= 0
G
(,)
, b

= x
(,,,,,z)
, where z

is the root of T (without loss

of generality there is a root). Dene M
1
= (M, a

), M
2
= (M, b

).
LABEL Sec4-
Claim-Equiv)
Claim 5.7. M
1
, M
2
are EF

T
equivalent.
We describe a winning strategy for ISO - this is very similar to the proof
of Claim 3.3, so we omit the details. We use the denitions in Denition Sec2-Claim-
Equiv-Models) 3.4. In every stage of the game ISO will choose a function g

such that:
Sec2-Def-W)
1. g
0
= .
2. g

J
i()+1
.
3. < g

W
g

.
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


EXISTENCE OF EF-EQUIVALENT NON ISOMORPHIC MODELS 21
4. If in stage AIS chose the sets A
1
, A
2
, then for each s I, if
G
s
(A
1
A
2
) ,= then (s) Dom(g

).
Now if = +1 and in stage AIS chose the sets A
1
, A
2
and the node z

,
ISO will dene h

= h
g
and then dene f

by
1. Dom(f

) =

G
s
: (s) Dom(g

),
2. for each s such that G
s
Dom(f

),
f

(0
Gs
) = x
t
, where t = t(s, g

, h

, z

).
LABEL Sec4-
Claim-Not-
Isomorphic)
Claim 5.8. M
1
, M
2
are not isomorphic.
Proof. It is enough to show that M is rigid. The proof is very similar to
the proof of Claim 4.10. Assume toward contradiction that f ,= id is an Sec3-Claim-Not-
Iso) automorphism of M. Denote
W
s
= t J
s
: x
t
is in the reduced representation of c
s
.
Since f ,= id there is s

= (u

) such that W
s
,= .
Case (*1): We can nd s

, t

: < ) such that

1. s

J, s

= (u

);
2. t

W
s

;
3. h
t

) =
i(

)+1
;
4. if < < , then i(

) < i(

).
In this case, note that the level of z
t

must be
i(

)
.
Case (*2): We cannot nd such a sequence. Therefore, for every large enough
i < , for every such that i() = i, for s() = (u

), for every
t W
s()
, we have h
t
() <
i+1
. Choose i

which satises this and

i
> .
We can nd t

, s

: <
i

+1
) such that
1. s

I, t

W
s

;
2. i(

) = i

;
3. if < , then h
t

) <

(< h
t
(

) ).
Since
i

+1
= cf(
i

+1
) >
+
i
and for every we have g
t

(x)
+
i
(this is
by Denition 5.52.(e)viii.), we may assume that g
t

) is constant. Sec4-Def-
Parameter) Now, in both cases, we proceed in a similar way to the proof of Claim
4.10. Using Claim 5.4, we choose A such that
Sec3-Claim-Not-
Iso)
Sec4-Claim-
exists-inner-
model)
1. [A[ =
0
;
2. W
s

: A) M;
3. if we are in case (*1), then z
t

: A) is an antichain in
T
(we
can demand this because in Case (*1) the levels of the z
t

s are not
bounded in - see Claim 5.4).
Sec4-Claim-
exists-inner-
model)
Dene s
+
I by s
+
= (,

g
t
, h
t
: t W
s

, A).
LABEL Sec4-
Claim-unique-
projection)
Claim 5.9. For every A the following holds: If r J
s
+, t W
s

, (r, t)
T, then
1. if (g, h) is a witness for r, then g
t
g and h
t
h;
2. if t ,= t

J
s

, then (r, t

) / T.
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


22 CHANOCH HAVLIN AND SAHARON SHELAH
Proof. See the proof of Claim 4.11.
5.9
Sec3-Claim-
unique-projection)
Sec4-Claim-
unique-projection)
LABEL Sec4-
Claim-exists-
projection)
Claim 5.10. For every A there is r W
s
+ such that (r, t

) T.
Proof. See the proof of Claim 4.12.
5.10
Sec3-Claim-
exists-projection)
Sec4-Claim-
exists-projection)
Now, using Claim 5.10, we choose for each A an r

W
s
+ such that
Sec4-Claim-
exists-projection)
(t

, r

) T.
LABEL Sec4-
claim-r-ies-
distinct)
Claim 5.11. If < , then r

,= r

.
Proof. If we are in Case (*1): z
t

, z
t
are not comparable. But z
r

T
z
t

because they are comparable and z

r

is on greater level, since that level

is determined by Denition 5.5,2.(d). By the same argument, z
r

T
z
t
.
Sec4-Def-
Parameter)
Therefore, z
r
, z
r

are not comparable, so r

,= r

.
If we are in Case (*2): Assume towards contradiction that r = r

= r

.
Let (g, h) be a witness for r. Then by Claim 5.9, g
t

, g
t
g and h
t

, h
t
h.
Sec4-Claim-
unique-projection)
Since we are in Case (*2) we get that
g(

) = g(

) and h(

) <

< h(

)
which contradicts the denition of a witness (see Denition 5.5,2.(e)).
5.11
Sec4-Def-
Parameter)
Sec4-claim-r-ies-
distinct)
We got that W
s
+ is innite - contradiction. Therefore, M must be rigid,
which proves Claim 5.8.
5.8
Sec4-Claim-Not-
Isomorphic)
Sec4-Claim-Not-
Isomorphic)
The proof of Theorem 5.3 is now complete
5.8
Sec4-Theorem-
EF*-T-Equiv)
Sec4-Claim-Not-
Isomorphic)
8
6
6


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







m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
3
-
0
4


EXISTENCE OF EF-EQUIVALENT NON ISOMORPHIC MODELS 23
References
[1] C.C Chang, Some remarks on the model theory of innitary languages, in The syntax
and semantics of innitary languages, Lecture notes in Mathematics, 72, J. Barwise
ed. (Springer, Berlin, 1968) pp. 36-63
[2] T.Hyttinen and H. Tuuri, Constructing strongly equivalent nonisomorphic models for
unstable theories, Annals of pure And Applied Logic 52(1991) pp.203-248
[3] Hodges, Wilfrid Model theory. Encyclopedia of Mathematics and its Applications, 42.
Cambridge University Press,Cambridge, 1993. ISBN: 0-521-30442-3 (Reviewer: J. M.
Plotkin)
[4] S.Shelah A long EF equivalence non isomorphic models - to appear (SH 836 in Shelah
archive)
[5] S.Shelah, The generalized continum hypothesis revisited, Israel J. Math 116 (2000) pp.
285-321
[6] S.Shelah, Existence of many L
,
non isomorphic models of power for singular
with

= , Notre Dame J. Formal Logic 25(1984) pp. 97-104

[7] S.Shelah, Existence of many L
,
non isomorphic models of T of power , Annals of
Pure and Applied Logic 34 (1987) pp. 291-310
[8] S.Shelah, Classication Theory, Stud. Logic Found. Math. 92 (North Holland, Ams-
terdam, 2nd rev. ed. 1990)
[9] S.Shelah, Cardinal Arithmetic, Oxford Logic Guide 29, Oxford University Press, 1994.
[10] D.S. Scott, Logic with denumerably long formulas and nite strings of quantiers
in The theory of models, J.W.Adisson, L.Henkin and A.Tarsky , eds. (North holland,
Amsterdam, 1965) pp. 329-341
[11] J. Vaananen, Games and trees in innitary logic: A survay, in Quantiers, M. Kryn-
icki, M. Mostowski and L. Szczerba, eds. (Kluwer, 1995) pp. 105-138.
References
[Vaa95] Jouko Vaananen. Games and trees in innitary logic: A survey. In M. Mostowski
M. Krynicki and L. Szczerba, editors, Quantiers, pages 105138. Kluwer, 1995.
Institute of Mathematics The Hebrew University of Jerusalem Jerusalem
91904, Israel
E-mail address: chanoch.havlin@gmail.com
Institute of Mathematics, The Hebrew University of Jerusalem, Jerusalem
91904, Israel, and, Department of Mathematics, Rutgers University, New
Brunswick, NJ 08854, USA
E-mail address: shelah@math.huji.ac.il
URL: http://www.math.rutgers.edu/~shelah