Documente Academic
Documente Profesional
Documente Cultură
6
9
r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
8
m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
0
8
On regular reduced products
Juliette Kennedy
Department of Mathematics
University of Helsinki
Helsinki, Finland
Saharon Shelah
Institute of Mathematics
Hebrew University
Jerusalem, Israel
October 6, 2003
Abstract
Assume
0
,
1
,
+
. Assume M is a model of a rst order
theory T of cardinality at most
+
in a vocabulary L(T) of cardinality
. Let N be a model with the same vocabulary. Let be a set of
rst order formulas in L(T) and let D be a regular lter on . Then
M is -embeddable into the reduced power N
/D is
++
-universal. Our second result is as follows: For
i < let M
i
and N
i
be elementarily equivalent models of a vocabulary
which has has cardinality . Suppose D is a regular lter on and
0
,
1
,
+
holds. We show that then the second player has
a winning strategy in the Ehrenfeucht-Fraisse game of length
+
on
i
M
i
/D and
i
N
i
/D. This yields the following corollary: Assume
GCH and regular (or just
0
,
1
,
+
and 2
=
+
). For L,
M
i
and N
i
be as above, if D is a regular lter on , then
i
M
i
/D
=
i
N
i
/D.
This paper was written while the authors were guests of the Mittag-Leer Institute,
Djursholm, Sweden. The authors are grateful to the Institute for its support.
/F is
1
-saturated and hence
2
-universal ([6]). This
was generalized by Shelah in [10] to any lter F on for which B
/F is
1
-saturated, where B is the two element Boolean algebra, and in [8] to all
regular lters on . In the rst part of this paper we use the combinatorial
principle 2
b
/D is
++
-universal.
The second part of this paper addresses Problem 18 in [3], which asks if
it is true that if D is a regular ultralter over , then for all elementarily
equivalent models M and N of cardinality in a vocabulary of cardinality
, the ultrapowers M
/D and N
=
+
. Benda [1] weakened good to contains
a good lter. We prove the claim in full generality, assuming 2
=
+
and
0
,
1
) ,
+
).
Regarding our assumption
0
,
1
) ,
+
), by Changs Two-Cardinal
Theorem ([2])
0
,
1
) ,
+
) is a consequence of =
<
. So our
Theorem 2 settles Conjecture 19 of [3], and Theorem 13 settles Conjecture 18
of [3], under GCH for regular. For singular strong limit cardinals
0
,
1
)
,
+
) follows from 2
0
,
1
) ,
1
,
2
), so our assumption is independent of ZFC.
2 Universality
Denition 1 Suppose is a set of rst order formulas of vocabulary L.
The set of -existential formulas is the set of formulas of the form
x
1
...x
n
(
1
...
n
),
where each
i
is in . The set of weakly -existential formulas is the set
of formulas of the above form, where each
i
is in or is the negation of a
formula in . If M and N are L-structures and h : M N, we say that h
is a -homomorphism if h preserves the truth of -formulas. If h preserves
also the truth of negations of -formulas, it is called a -embedding.
2
7
6
9
r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
8
m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
0
8
Theorem 2 Assume
0
,
1
) ,
+
). Let M be a model of a rst order
theory T of cardinality at most
+
, in a language L of cardinality and
let N be a model with the same vocabulary. Let be a set of rst order
formulas in L and let D be a regular lter on . We assume that every
weakly -existential sentence true in M is true also in N. Then there is a
-embedding of M into the reduced power N
/D.
By letting be the set of all rst order sentences, we get from Theorem 2
and Los Lemma:
Corollary 3 Assume
0
,
1
) ,
+
). If M is a model with vocabulary
, and D is a regular ultralter over , then M
/D is
++
-universal,
i.e. if M
is of cardinality
+
, and M
M, then M
is elementarily
embeddable into the ultrapower M
/D.
We can replace weakly -existential by -existential in the Theorem,
if we only want a -homomorphism.
The proof of Theorem 2 is an induction over and
+
respectively, as
follows. Suppose M = a
: <
+
. We associate to each <
+
nite
sets u
i
, i < , and represent the formula set as a union of nite sets
i
.
At stage i, for each <
+
we consider the
i
-type of the elements a
of the
model whose indices lie in the set u
i
, <
+
. This will yield a witness f
(i)
in N at stage i, . Our embedding is then given by a
i
and integers n
i
for each <
+
and i < such that for each i,
(i) [u
i
[ < n
i
<
(ii) u
i
(iii) Let B be a nite set of ordinals and let be such that B <
+
.
Then i : B u
i
D
(iv) Coherency: u
i
u
i
= u
i
Assuming the lemma, and letting M = a
: <
+
we now dene, for
each , a function f
: N.
3
7
6
9
r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
8
m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
0
8
Let =
is nite. Let
i
=
: w
i
, and let u
i
, n
i
be as in the lemma.
We dene a sequence of formulas essential to the proof: suppose <
+
and i < . Let m
i
= [u
i
[ and let
u
i
=
,i,0
, ...,
,i,m
i
1
be an increasing enumeration of u
i
. (We adopt henceforth the convention
that any enumeration of u
i
that is given is an increasing enumeration.)
Let
i
be the
i
-type of the tuple a
,i,0
, ..., a
,i,m
i
1
, a
) in M. (So ev-
ery (x
,i,0
, ..., x
,i,m
i
1
, x
)
i
or its negation occurs as a conjunct of
i
, according to whether (a
,i,0
, ..., a
,i,m
i
1
, a
) or (a
,i,0
, ..., a
,i,m
i
1
, a
)
holds in M.) We dene the formula
i
for each i by downward induction on
m
i
as follows:
Case 1: m
i
= n
i
. Let
i
=
i
.
Case 2: m
i
< n
i
. Let
i
be the conjunction of
i
and all formulas of the
form x
m
i
(x
0
, ..., x
m
i
1
, x
m
i
), where satises u
i
= u
i
and hence
m
i
= m
i
+ 1.
An easy induction shows that for a xed i < , the cardinality of the set
i
: <
+
is nite, using n
i
.
Let i < be xed. We dene f
< and u
i
= r
1
, ..., r
k
, then N [=
i
(f
1
(i), ..., f
k
(i), f
(i)).
To dene f
i
.
Case 1.1: n
i
= 0. Then
i
is the
i
type of the element a
. But then
M [=
i
(a
)
M [= x
0
i
(x
0
)
N [= x
0
i
(x
0
),
4
7
6
9
r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
8
m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
0
8
where the last implication follows from the assumption that N satises the
weakly -existential formulas holding in M. Now choose an element b N
to witness this formula and set f
(i) = b.
Case 1.2: n
i
> 0. Let u
i
=
1
, . . . ,
m
i
. Since m
i
= n
i
, the formula
i
is the
i
-type of the elements
1
, . . . ,
m
i
. By assumption =
m
i
is the
maximum element of u
i
. Thus by coherency, u
i
= u
i
=
1
, . . . ,
m
i
1
.
Since < , we know by the induction hypothesis that
N [=
i
(f
1
(i), . . . , f
i
(i)).
By the formula construction
i
contains the formula x
m
i
(x
1
, . . . , x
m
i
),
since u
i
= u
i
and since m
i
< n
i
. Thus
N [= x
m
i
+1
i
(f
1
(i), . . . , f
i
(i), x
m
i
+1
).
As before choose an element b N to witness this formula and set f
(i) = b.
Case 2: m
i
< n
i
. Let u
i
=
1
, . . . ,
m
i
. We have that M [=
i
(a
1
, . . . , a
i
, a
),
and therefore M [= x
m
i
+1
i
(a
1
, . . . , a
i
, x
m
i
+1
). Let = max(u
i
)=
m
i
.
By coherency, u
i
= u
i
and therefore since < by the induction hy-
pothesis we have that
N [=
i
(f
1
(i), . . . , f
i
1
(i), f
(i)).
But then as in case 1.2 we can infer that
N [= x
m
i
+1
i
(f
1
(i), . . . , f
i
(i), x
m
i
+1
).
As in case 1 choose an element b N to witness this formula and set f
(i) = b.
It remains to be shown that the mapping a
1
, . . . , a
k
) i : N [= (f
1
(i), . . . , f
k
(i)) D.
So let such a be given, and suppose M [= (a
1
, . . . , a
k
). Let I
= i :
N [= (f
1
(i), . . . , f
k
(i)). We wish to show that I
. Let <
+
be such
that
1
, ...,
n
. By Lemma 4 condition (iii), i :
1
, ...,
n
u
i
D.
So it suces to show
A
i :
1
, ...,
n
u
i
I
.
Let i A
such that
1
, ...,
n
u
i
. By the denition of
i
we know that
N [=
i
(f
1
(i), . . . , f
k
(i)). But the
i
-type of the tuple
1
, . . . ,
k
) occurs as
a conjunct of
i
, and therefore N [= (f
1
(i), . . . , f
k
(i)) 2
3 Proof of Lemma 4
We now prove Lemma 4. We rst prove a weaker version in which the lter
is not given in advance:
Lemma 5 Assume
0
,
1
) ,
+
). There exist sets u
i
: <
+
, i <
cof()), integers n
i
and a regular lter D on , generated by sets, such
that (i)-(iv) of Lemma 4 hold.
Proof. By [11, Proposition 5.1, p. 149] the assumption
0
,
1
) ,
+
)
is equivalent to:
2
b
: There is a
+
-like linear order L, sets C
a
: a L, < cf()), equiva-
lence relations E
a,b
: < , a L, b
L) such that
(i)
a
= b : b <
L
a (an increasing union in ).
(ii) If b C
a
, then C
b
= c C
a
: c <
L
b.
(iii) E
renes E
.
(v) If aE
b, then f
a,b
is an order-preserving one to one mapping from
C
a
onto C
b
such that for d C
a
, dE
a,b
(d).
(vi) If < < cf() and aE
b, then f
a,b
f
a,b
.
(vii) If f
a,b
(a
1
) = b
1
, then f
a
1
,b
1
f
a,b
.
(viii) If a C
b
then E
(a, b).
6
7
6
9
r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
8
m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
0
8
This is not quite enough to prove Lemma 5, so we have to work a little
more. Let
= a/E
: a L.
We assume, for simplicity, that ,= implies
= . Dene for
t
1
, t
2
:
t
1
<
t
2
(a
1
t
1
)(a
2
t
2
)(a
1
C
a
2
).
Proposition 6
, <
t
2
<
t
3
implies t
1
<
t
3
, and (b) t
1
<
t
3
and t
2
<
t
3
implies t
1
<
t
2
or t
2
<
t
1
or t
1
= t
2
. For the rst, t
1
<
t
2
implies there exists a
1
t
1
and a
2
t
2
such that a
1
C
a
2
. Similarly t
2
<
t
3
implies there exists b
2
t
2
and b
3
t
3
such that b
2
C
b
3
. Now a
2
E
b
2
and
hence we have the order preserving map f
a
2
,b
2
from C
a
2
onto C
b
2
. Recalling
a
1
C
a
2
, let f
a
2
,b
2
(a
1
) = b
1
. Then by (vi), a
1
E
b
1
and hence b
1
t
1
. But
then b
1
C
b
2
implies b
1
C
b
3
, by coherence and the fact that b
2
C
b
3
. But
then it follows that t
1
<
t
3
.
Now assume t
1
<
t
3
and t
2
<
t
3
. Let a
1
t
1
and a
3
t
3
be such that
a
1
C
a
3
, and similarly let b
2
and b
3
be such that b
2
C
b
3
. a
3
E
b
3
implies we
have the order preserving map f
a
3
,b
3
from C
a
3
to C
b
3
. Letting f
a
3
,b
3
(a
1
) = b
1
,
we see that b
1
C
b
3
. If b
1
<
L
b
2
, then we have C
b
2
= C
b
3
c[c < b
2
which
implies b
1
C
b
2
, since, as f
a
3
,b
3
is order preserving, b
1
<
L
b
2
. Thus t
1
<
t
2
.
The case b
2
<
L
b
1
is proved similarly, and b
1
= b
2
is trivial. 2
For a <
L
b let
(a, b) = min : a C
b
.
Denoting (a, b) by , let
tp(a, b) = a/E
, b/E
).
If a
1
<
L
... <
L
a
n
, let
tp(a
1
, ..., a
n
)) = l, m, tp(a
l
, a
m
))[1 l < m n
and
= tp(a) : a
<
L.
For t = tp(a), a
n
L we use n
t
to denote the length of a.
7
7
6
9
r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
8
m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
0
8
Proposition 7 If a
0
<
L
... <
L
a
n
, then
max(a
l
, a
m
) : 0 l < m n = max(a
l
, a
n
) : 0 l < n.
Proof. Clearly the right hand side is the left hand side. To show the
left hand side is the right hand side, let l < m < n be arbitrary. If
(a
l
, a
n
) (a
m
, a
n
), then (a
l
, a
m
) (a
m
, a
n
). On the other hand, if
(a
l
, a
n
) > (a
m
, a
n
), then (a
l
, a
m
) (a
l
, a
n
). In either case (a
l
, a
m
)
max(a
k
, a
n
) : 0 k < n. 2
Let us denote max(a
l
, a
n
) : 0 l < n by (a). We dene on a
two-place relation
as follows:
t
1
<
t
2
if there exists a tuple a
0
, . . . a
nt
2
1
) realizing t
2
such that some subsequence
of the tuple realizes t
1
.
Clearly, ,
a and
b
k
2
E
a, k
1
, k
2
n 1. Without loss of generality, assume b
k
1
< b
k
2
. Since
E
is an equivalence relation, b
k
2
E
b
k
1
and thus we have an order preserving
map f
b
k
2
,b
k
1
from C
b
k
2
to C
b
k
1
. Also b
k
1
C
b
k
2
, by the denition of and by
coherence, and therefore f
b
k
2
,b
k
1
(b
k
1
)E
b
k
1
. But this contradicts (viii), since
f
b
k
2
,b
k
1
(b
k
1
) C
b
k
1
. 2
Denition 9 For t , t = tp(b
0
, . . . b
n1
) and a L suppose there exists
k < n such that b
k
E
(b
0
,...,b
n1
)
a. Then let u
a
t
= f
(b
0
,...,b
n1
)
a,b
k
(b
l
) : l < k
Otherwise, let u
a
t
= .
Finally, let D be the lter on generated by the sets
t
= t : t
<
L
t.
We can now see that the sets u
a
t
, the numbers n
t
and the lter D satisfy
conditions (i)-(iv) of Lemma 4 with L instead of
+
: Conditions (i) and (ii)
8
7
6
9
r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
8
m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
0
8
are trivial in this case. Condition (iii) is veried as follows: Suppose B is
nite. Let a L be such that (x B)(x <
L
a). Let a enumerate B a
in increasing order and let t
= tp(a). Clearly
t
t
B u
a
t
.
Condition (iv) follows directly from Denition 9 and Proposition 8.
To get the Lemma on
+
we observe that since L is
+
-like, we can assume
that
+
, <) is a submodel of L, <
L
). Then we dene v
t
= u
t
: < .
Conditions (i)-(iv) of Lemma 5 are still satised. Also having D a lter of
instead of is immaterial as [[ = . 2
Now back to the proof of Lemma 4. Suppose u
i
, n
i
and D are as in
Lemma 5, and suppose D
: <
be a family of sets witnessing the regularity of D
, and let Z
: < be
the family generating D. We dene a function h : as follows. Suppose
i < . Then let
h(i)
[i A
.
Now dene v
= u
h()
. Dene also n
= n
h()
. Now the sets v
and the
numbers n
i
C
i
a
= a.
(iii) If b C
i
a
, then C
i
b
= C
i
a
b.
(iv) supotp(C
i
a
) : a <
+
< .
Thus it suces to prove:
9
7
6
9
r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
8
m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
0
8
Proposition 10 Suppose the sets u
i
and the lter D are as given by Lemma 5
and is a limit cardinal. Then o
holds.
Proof. Suppose / = A
: < cf())
satisfy:
(a) [
(b)
, then there is
such that
A
= A
1
... A
n
.
The sequence
= < : (
)(A
i : u
i
).
Lemma 11 (1) V
, and
: < cf() = .
(2) If V
, then V
= V
.
Proof. (1) is a direct consequence of the denitions. (2) follows from the
respective property of the sets u
i
. 2
Lemma 12 supotp(V
) : <
+
+
.
Proof. By the previous Lemma, [V
. Therefore otp(V
) <
+
and
the claim follows. 2
The proof of the proposition is complete: (i)-(iii) follows from Lemma 11,
(iv) follows from Lemma 12 and the assumption that is a limit cardinal. 2
More equivalent conditions for the case singular strong limit, D a regular
ultralter on , are under preparation.
10
7
6
9
r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
8
m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
0
8
5 Ehrenfeucht-Frasse-games
Let M and N be two rst order structures of the same vocabulary L. All
vocabularies are assumed to be relational. The Ehrenfeucht-Frasse-game of
length of M and N denoted by EFG
, y
. Eventually a sequence
(x
, y
) : < ) has been played. The rules of the game say that both
players have to play elements of M N. Moreover, if I plays his x
in M
(N), then II has to play his y
, y
) : < )
determines a relation MN. Player II wins this round of the game if
is a partial isomorphism. Otherwise I wins. The notion of winning strategy
is dened in the usual manner. We say that a player wins EFG
if he has a
winning strategy in EFG
.
Note that if II has a winning strategy in EFG
on M and N, where M
and N are of size [[, then M
= N.
Assume L is of cardinality and for each i < let M
i
and N
i
are
elementarily equivalent L-structures. Shelah proved in [12] that if D is a
regular lter on , then Player II has a winning strategy in the game EFG
on
i
M
i
/D and
i
N
i
/D for each <
+
. We show that under a stronger
assumption, II has a winning strategy even in the game EFG
+. This makes
a big dierence because, assuming the models M
i
and N
i
are of size
+
,
2
=
+
, and the models
i
M
i
/D and
i
N
i
/D are of size
+
. Then by
the remark above, if II has a winning strategy in EFG
+ on
i
M
i
/D and
i
N
i
/D.
Proof. We use Lemma 4. If i < , then, since M
i
and N
i
are elementarily
equivalent, Player II has a winning strategy
i
in the game EFG
n
i
on M
i
and
N
i
. We will use the set u
i
to put these short winning strategies together into
one long winning strategy.
11
7
6
9
r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
8
m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
0
8
A good position is a sequence (f
, g
i
M
i
, g
i
N
i
, and if i < , then (f
(i), g
(i)) :
u
i
) is a play according to
i
.
Note that in a good position the equivalence classes of the functions f
and
g
so that (f
, g
(i), g
(i)) : u
i
). We know that this is a play according to the strategy
i
and [u
i
[ < n
i
. Thus we can play one more move in EF
n
i
on M
i
and
N
i
with player I playing f
(i). Let g
. We have
showed that II can maintain a good position. 2
Corollary 14 Assume GCH and regular (or just
0
,
1
) ,
+
) and
2
=
+
). Let L be a vocabulary of cardinality and for each i < let M
i
and N
i
be two elementarily equivalent L-structures. If D is a regular lter
on , then
i
M
i
/D
=
i
N
i
/D.
References
[1] M. Benda, On reduced products and lters. Ann.Math.Logic 4 (1972),
1-29.
[2] C. C. Chang, , A note on the two cardinal problem, Proc. Amer. Math.
Soc., 16, 1965, 11481155,
[3] C.C. Chang and J.Keisler, Model Theory, North-Holland.
[4] T. Hyttinen, On -complete reduced products, Arch. Math. Logic,
Archive for Mathematical Logic, 31, 1992, 3, 193199
[5] R. Jensen, The ne structure of the constructible hierarchy, With a
section by Jack Silver, Ann. Math. Logic, 4, 1972, 229308
[6] B. J onsson and P. Olin, Almost direct products and saturation, Com-
positio Math., 20, 1968, 125132
12
7
6
9
r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
5
-
0
8
m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
5
-
0
8
[7] J. Keisler, Ultraproducts and saturated models. Nederl.Akad.Wetensch.
Proc. Ser. A 67 (=Indag. Math. 26) (1964), 178-186.
[8] J. Kennedy and S. Shelah, On embedding models of arithmetic of car-
dinality
1
into reduced powers, to appear.
[9] W. Mitchell, Aronszajn trees and the independence of the transfer prop-
erty, Ann. Math. Logic, 5, 1972/73, 2146
[10] S. Shelah, For what lters is every reduced product saturated?, Israel J.
Math., 12, 1972, 2331
[11] S. Shelah, Gap 1 two-cardinal principles and the omitting types theo-
rem for L(Q). Israel Journal of Mathematics vol 65 no. 2, 1989, 133152.
[12] S. Shelah, Classication theory and the number of non-isomorphic
models, Second, North-Holland Publishing Co., Amsterdam, 1990,
xxxiv+705
13