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

0
1

1
2
m
o
d
i
f
i
e
d
:
2
0
0
2

0
1

1
2
On equivalence relations second order denable over
H()
Saharon Shelah
October 6, 2003
Abstract
Let be an uncountable regular cardinal. Call an equivalence re
lation on functions from into 2 second order denable over H()
if there exists a second order sentence and a parameter P H()
such that functions f and g from into 2 are equivalent i the struc
ture H(), , P, f, g satises . The possible numbers of equivalence
classes of second order denable equivalence relations contains all the
nonzero cardinals at most
+
. Additionally, the possibilities are closed
under unions and products of at most cardinals. We prove that these
are the only restrictions: Assuming that GCH holds and is a cardinal
with
= , and
for all cardinals , the number of equivalence classes of some second
order denable equivalence relation on functions from into 2 is i
is in , where is any prearranged subset of such that 0 ,
contains all the nonzero cardinals
+
, and is closed under unions
and products of at most cardinals.
1
1 Introduction
We deal with equivalence relations which are second order denable over
H(), where is an uncountable regular cardinal. We show that it is possible
1
1
denable over H() with equivalence classes. There is also a
1
1

equivalence relation having
+
classes (Lemma 3.2). Furthermore, by a
simple coding, the possible numbers of equivalence classes of
1
1
equivalence
relations are closed under unions of length and products of length < .
In other words, assuming that and
i
, i < , are cardinals such
that for each i < , there is a
1
1
equivalence relation having
i
equivalence
classes, there exists a
1
1
equivalence relation having
i<
i
equivalence
classes. Similarly, if < , there exists also a
1
1
equivalence relation with
card(
i<
i
) equivalence classes (Lemma 3.4).
What are the possible numbers of equivalence classes between
+
and 2
?
The existence of a tree T H() with branches of length through it
implies that there is a
1
1
equivalence relation having equivalence classes
(Lemma 3.2). Therefore, existence of a Kurepa tree of height with more
than
+
and less than 2
>
+
, there is no denable equivalence relation having
dierent equivalence classes when
+
< < 2
=
+
. Let
>
+
be a cardinal with
, or otherwise, 2) the number of equivalence classes really depends on
trees, not less than . The latter case is the most dicult and it is presented
in Subsection 4.4. There we notice that the xed equivalence relation must
have classes, where is a union of products of cardinals in
.
In Section 5 we present some remarks.
2 Adding Kurepa trees
Throughout of the paper we assume that is an uncountable regular cardinal
and
<
= . For sets X and Y we denote the set of all functions from X
into Y by
X
Y . For a cardinal , we let [X]
as follows.
It consists of all pairs p =
T
p
, b
p

p
)
_
, where
for some < , T
p
is a subset of 
2 and < such that it
is of cardinality < and closed under restriction;
p
is a subset of having cardinality < and each b
p
is an branch
trough T
p
when T
p
is ordered by the inclusion.
For all p, q P
, we dene that q p if
T
q
is an endextension of T
p
;
p
q
;
for every
p
, b
q
is an extension of b
p
.
Fact 2.2
(a) P
() = b
p
).
Proof. The idea of the proof is the same as in [Jec71]. Suppose p
0
, q
0
) is a
condition in P
Q and
t is a name such that
p
0
, q
0
)
P
t is a branch through
T
G
and
t ,
b
G
 < .
Since 1
P
Q
is a regular cardinal, it follows that every condition be
low p
0
, q
0
) forces that for all X []
<
and < , there is > with
t() ,
b
G
()  X.
Let
0
be the height of T
p
0
. Choose conditions p
n
, q
n
) from P
Q and
ordinals
n
, 1 < n < , so that for every n < , the height of the tree T
p
n+1
is greater than
n
, p
n+1
, q
n+1
) p
n
, q
n
), and
(A) p
n+1
, q
n+1
)
(P
Q)
t(
n+1
) ,
b
G
(
n+1
) 
pn
.
Dene r to be the condition in P
satisfying T
r
=
n<
T
pn
,
r
=
n<
pn
,
and for every
r
, b
r
=
n(m)
b
pn
(),
r
, by f
,
and dene r
to be the condition in P
with T
r
= T
r
f

r
,
=
r
, and for every
r
and ,
b
r
() =
_
b
r
() if < ;
f
if = .
Now r
,
r
.
Since r
forces
Q to be closed and q
n
 n < ) to be a decreasing sequence
of conditions, there is q
so that r
, q
) p
n
, q
n
) for every n < . Since
r
, q
) forces that
t() f

r
, there are
r
and a condition
5
7
1
9
r
e
v
i
s
i
o
n
:
2
0
0
2

0
1

1
2
m
o
d
i
f
i
e
d
:
2
0
0
2

0
1

1
2
r
, q
) r
, q
) in P
Q forcing that
t() = f
. However, if n is the
smallest index with
pn
, then r
, q
) forces that
t(
n+1
) = f
n+1
= b
r
(
n+1
) = b
r
(
n+1
) =
b
G
(
n+1
),
contrary to (A). 2.3
Denition 2.4 Suppose >
+
is a cardinal with
= . Let =

< ) be a xed sequence of cardinals such that <
= has cardinality . We
dene P( ) to be the product of P
forcings:
P( ) is the set of all functions p such that dom(p) is a subset of with
cardinality < , and for every dom(p), p() is a condition in P
;
the order of P( ) is dened coordinate wise, i.e., for p, q P( ), q p
if dom(p) dom(q) and for every dom(p), q() p().
The weakest condition in P( ) is the empty function, denoted by 1. For each
p P( ) and dom(p), we let the condition p() be the pair T
p
, b
p
,

p
.
Fact 2.5
(a) The forcing P( ) is closed and it has
+
c.c..
(b) Suppose G is a P( )generic set over V. In V[G], for every < , the
branches through the tree T
G
=
pG
T
p
are b
G
,
 <
, where
each b
G
,
is the function
_
b
p
,
 p G, dom(p) and
p
.
Proof. (b) Since 1
P( (+1))
P( ( ( + 1))) is closed, the claim
follows from Lemma 2.3.
Denition 2.6 For all P( )names , dene that
=
_
p
 condition p appears in .
Let
1st
denote the set  , )
and
.
6
7
1
9
r
e
v
i
s
i
o
n
:
2
0
0
2

0
1

1
2
m
o
d
i
f
i
e
d
:
2
0
0
2

0
1

1
2
Denition 2.7 We denote by Sqs( ) the set of all sequences z = z

Z) such that Z and for each Z, z
is a subset of
of cardinality
at least . In order to keep our notation coherent, let
z
be a shorthand for
the set
Z
z
 Y )
in Sqs( ). In addition, there exists an injective function
1st
from Y into
and injective functions
from y
into
()).
For every z Sqs( ), Mps( z) is the collection
_
Mps( )  dom()
z
_
.
Denition 2.10 For every p P( ) and Mps( ) with
p
dom(),
we let (p) denote the condition q in P( ) for which
dom(q) =
1st
[dom(p)],
for every dom(q), T
q
= T
p
and
q
[
p
], where = (
1st
)
1
();
7
7
1
9
r
e
v
i
s
i
o
n
:
2
0
0
2

0
1

1
2
m
o
d
i
f
i
e
d
:
2
0
0
2

0
1

1
2
for every , )
q
, b
q
,
= b
p
,
, where , ) =
1
, ).
When is a P( z)name and a mapping in Mps( ) with
dom(),
() denotes the P( z)name which is result of recursively replacing every
condition p in with (p), i.e.,
() = (), (p))  , p) .
Analogously, for sequences z = z
 Z) with
z
dom(), we let ( z)
denote the sequence z
 Z
), where Z
=
1st
[Z] and for each Z
,
z
[z
] with = (
1st
)
1
().
Fact 2.11 For every subforcing P( z) and Mps( z) with dom() =
z
,
the mapping p (p) is an isomorphism between P( z) and P(( z)).
Suppose P( z) is a subforcing of P( ). The isomorphism determined by some
Mps( z) is denoted by . It follows that if dom() =
z
, p P( z),
(x
1
, . . . , x
n
) with n < is a formula, and
1
, . . . ,
n
are P( z)names then
(2.1) p
P( z)
(
1
, . . . ,
n
) i (p)
P(( z))
( (
1
), . . . , (
n
)).
Particularly, a mapping in Mps( z) determines an automorphism of P( z)
when
1st
is a permutation of Z and each
is a bijection from z
onto
z
1st
()
.
3 Basic facts on
1
1
equivalence relations
Recall that we assume to be an uncountable regular cardinal. We let H()
denote the set of all sets having transitive closure of cardinality < .
Denition 3.1 We say that denes an equivalence relation
,R
on
2
with a parameter R H() when
is a second order sentence in the vocabulary consisting of , one
unary relation symbol S
0
, and binary relation symbols S
1
and S
2
;
the following denition gives an equivalence relation on
2: for all
f, g
2
f
,R
g i H(), , R, f, g) [= ,
where R, f, and g are the interpretations of the symbols S
0
, S
1
, and
S
2
respectively.
8
7
1
9
r
e
v
i
s
i
o
n
:
2
0
0
2

0
1

1
2
m
o
d
i
f
i
e
d
:
2
0
0
2

0
1

1
2
An equivalence relation is
1
1
denable, if denes it and is of the form
X(S
0
, S
1
, S
2
, X), where X is the only second order variable appearing in
.
We abbreviate card
_
f/
,R
 f
2
_
by No(
,R
).
Lemma 3.2
(a) For every nonzero cardinal , 2
, there exists a
1
1
equivalence
relation
,R
on
2 with No(
,R
) = .
(b) There exists a
1
1
equivalence relation
,R
on
2 with No(
,R
) =
+
.
(c) If T is a tree with card(T) = , then there exists a
1
1
equivalence rela
tion
,R
on
2 with No(
,R
) = card(Br
(T)) + 1.
Proof. Let be a xed denable bijection from onto . For a binary
relation R, we denote the set ()  for some < , , 1) R by (R).
(a) In the cases , the parameter can code a list of nonequiv
alent functions. In the case No(
,R
) = 2
1
1
equivalence relation
,R
with No(
,R
) = .
Proof. The claim follows from Fact 2.5 together with Lemma 3.2.
In the next section we shall need the following properties of
1
1
equivalence
relations.
Lemma 3.4 Suppose and
i
, i < , are nonzero cardinals such that
i
denes a
1
1
equivalence relation on
2 with the parameter R
i
and it has
i
equivalence classes.
(a) There exists a
1
1
equivalence relation
,R
on
2 with No(
,R
) =
i<
i
.
(b) There exists a
1
1
equivalence relation
,R
on
2 with No(
,R
) =
card(
i<
i
).
Proof. Both of the claims are simple corollaries of the fact that there are a
parameter R H() and a formula (x) such that for all f, g, h
2
H(), , R, f, g, h) [= (i)
if, and only if,
H(), , R[i], f[i], g[i], h[i]) [=
i
,
where R[i], f[i], g[i], and h[i] are the i
th
parts of R, f, g, and h respectively,
in some denable coding. Furthermore R[i] = R
i
holds for every i < .
4 Possible numbers of equivalence classes
Our goal is to show the consistency of the claim: the closure under unions
and products in Lemma 3.4 are the only restrictions on the possible numbers
of equivalence classes of equivalence relations on 2
 <
;
if and
i
, i < , are cardinals in
, then both
i<
i
and
card(
i<
i
) are in
.
Theorem 1 Suppose that
is an uncountable cardinal with
<
= and 2
=
+
;
>
+
is a cardinal with
= ;
=
+
holds.
Then for every P( )generic set G, the extension V[G] satises that all car
dinals and conalities are preserved, there are no new sets of cardinality
< , 2
R,
is ;
(C) there is a second order sentence dening over H() an equivalence
relation
,R
on
2 with a parameter R such that the number of equiv
alence classes of
,R
is .
Remark. Because P( ) does not add new subsets of cardinality < , the
denition of
yields the same sets in the ground model and in the generic
extension.
The rest of this section is devoted to the proof of this theorem. Because of
Conclusion 3.3 and Lemma 3.4 it remains to show that if
R is a P( )name
for a subset of H() and is a second order sentence such that in every
generic extension, denes over H() an equivalence relation
,R
on 2
,
then
1
P( )
No(
,
R
)
.
11
7
1
9
r
e
v
i
s
i
o
n
:
2
0
0
2

0
1

1
2
m
o
d
i
f
i
e
d
:
2
0
0
2

0
1

1
2
By Fact 2.8(b), we may assume that
R is chosen so that for xed cardinal
the following holds:
(4.1) 1
P( )
No(
,
R
) = .
Without loss of generality the name
R has cardinality . Of course we may
assume that >
+
, otherwise the claim follows from Lemma 3.2.
We prove that has the desired form, i.e., it is in
. The rst thing to
show is that No(
,
R
) depends only on the coordinates appearing in
R
with
R
classes. From the choice of the small z it follows that the
number of nonisomorphic names in the xed list is at most
+
. Here an
isomorphism class of
which
are images of
s. In the beginning of
Subsection 4.2 this is explained more formally.
So we x a name
+
holds for every
+
and < , it follows
that the cardinality of the isomorphism class of
is a small product or a
small union of cardinals in
.
Secondly, we assume that the number of critical indices is . Our assumption
on cardinal arithmetic in the ground model yields that all the products of
less than critical cardinals has smaller cardinality than the cardinality
of the isomorphism class under consideration. (Note that by the assumption
<
= , the number of such products is .) So the cardinality of the xed
isomorphism class is at least the supremum of all this type of products. On
12
7
1
9
r
e
v
i
s
i
o
n
:
2
0
0
2

0
1

1
2
m
o
d
i
f
i
e
d
:
2
0
0
2

0
1

1
2
the other hand, the cardinality of the isomorphism class cannot exceed this
union. A reason for this is that if the cardinality of the isomorphism class
is even larger than the supremum of all small products, then there are, by
lemma, coherent names for nonequivalent functions in the isomorphism
class (Lemma 4.9). Roughly speaking, such coherent names can be copied
by automorphisms of P( z) (xing the name for a parameter). This yields
more than names for nonequivalent functions. To prove all the details, we
introduce some technical tools in the beginning of Subsection 4.4.
4.1 Choice of a small subforcing
In this subsection we prove that there is a subforcing P( z) of P( ) such that
the cardinality of P( z) is , there are at most
+
coordinates in P( z), and
already P( z) produces dierent equivalent classes of
,
R
.
As mentioned above, the rst lemma will play a central role in the proof of
the main lemma of this subsection, Lemma 4.3.
For a regular cardinal , let Col(
+
, ) denote the standard
+
closed forc
ing notion collapsing the cardinality of to
+
.
Lemma 4.2
2
Suppose z Sqs( ) is such that dom( z) and each element in
ran( z) have cardinality
+
.
(a) Assume that y z, , q P( y), and
1
, . . . ,
n
are P( y)names
with n < and q
P( y)
1
, . . . ,
n
H(). Then for all Col(
+
, )
generic lters K and for all second order sentences in vocabulary
, R
1
, . . . , R
n
:
_
q
P( y)
H(), ,
1
, . . . ,
n
) [=
_
V
i
_
q
P( y)
H(), ,
1
, . . . ,
n
) [=
_
V[K]
.
(b) Suppose is a P( z)name, 1
P( z)
H(), is a mapping in Mps( )
such that card() , ran()
z
, and is identity on dom()
.
Let p be a condition in P( ) and a P( )name such that p
P( )
2
The authors wish to thank the anonymous referee for pointing out that instead of
1
1
denable equivalence relations one can consider arbitrary second order denable equiv
alence relations.
13
7
1
9
r
e
v
i
s
i
o
n
:
2
0
0
2

0
1

1
2
m
o
d
i
f
i
e
d
:
2
0
0
2

0
1

1
2
H() (w.l.o.g. both and have cardinality ). Then there exists
in Mps( ) of cardinality extending such that the domain of
contains
p
, ran()
z
, is identity on
2)
V[G]
, and
second order sentences in vocabulary , R
1
, R
2
, R
3
:
_
H(), , R, f, g) [=
_
V[G]
i
_
H(), , R, f, g) [=
_
V[H]
.
Proof. (a) Recall that we assume
<
= and hence P( y) has
+
c.c. in
V. Since Col(
+
, ) is
+
closed, cardinals
+
are preserved, (P( y))
V
=
(P( y))
V[K]
, and (H(
+
))
V[K]
= (H(
+
))
V
. Moreover,
<
= holds in
V[K], and hence P( y) has
+
c.c. in V[K] too.
For any P( y)name and q P( y), if q
P( y)
H(), then there exists
a nice P( y)name such that card() and q
P( y)
= . So for any
P( y)name in V[K] and q P( y) with q
P( y)
H(), there exists a
P( y)name in V with q
P( y)
= .
By the preservation of
+
c.c., for any A P( y): A is an antichain in V i A
is an antichain in V[K]. But the denition of q
P( y)
H(), ,
1
, . . . ,
n
) [=
depends only on possible antichains of P( y) and possible nice names for
subsets of H(). Hence the claim follows from the fact that V and V[K]
have the same antichains and the same nice names for subsets of H().
(b) Suppose that (p
P( )
H(), , , ) [= )
V
holds. By (a), (p
P( )
H(), , , ) [= )
V[K]
holds. Inside the generic extension V[K] there
exists a mapping in Mps( ) extending such that
1st
is a bijection from
dom( ) = into dom( z) which is identity on
1st
. Moreover, for every
dom( z),
is a bijection from
into z
which is identity on
(all
the cardinals between
+
and
+
are collapsed to
+
). Then determines
an isomorphism between P( ) and P( z). Therefore, ((p)
P( z)
H(),
14
7
1
9
r
e
v
i
s
i
o
n
:
2
0
0
2

0
1

1
2
m
o
d
i
f
i
e
d
:
2
0
0
2

0
1

1
2
, (), ()) [= )
V[K]
holds. Choose so that card() and
p
f
g
.
Suppose p H is a condition forcing the right hand side to be true. By (b)
there is a condition q P( z) forcing the left hand side to be true and q is
determined by a map in Mps( ) which is identity on
f
g
. The
only small problem is that we should have q G. However, the set of qs
like that is predense below p (i.e. each r p is compatible with some q like
that). The reason for this is that for any r P( ) with r p, r and (r)
are compatible, provided that for every , )
r
, either (, ) = , ) or
(, ) ,
r
. So given arbitrary r p, both and in (b) can be chosen
so that r and (r) are compatible. Therefore there must be some r p and
xing
R,
f, and g such that (r) H. Hence (r) is in G. 4.2
Remark. Even though the use of Col(
+
, ) provided an easy proof of the
previous lemma, the same idea cannot be applied in the proof of the next
lemma: might be a singular cardinal of conality
0
and hence Col(, )
is not even
1
closed.
Lemma 4.3 Recall that we assume >
+
and (4.1) on page 12 holds,
i.e.,
1
P( )
No(
,
R
) = .
Suppose P( z) is a subforcing of P( ) such that
z = z
 Z);
Z is a subset of satisfying card(Z) =
+
and
R
1st
Z;
for each
R
1st
, z
if
, and otherwise, z
_
y [
+

y
_
.
Z
R
1st
is of cardinality
+
(follows from the choice of
R);
15
7
1
9
r
e
v
i
s
i
o
n
:
2
0
0
2

0
1

1
2
m
o
d
i
f
i
e
d
:
2
0
0
2

0
1

1
2
for every Z
R
1st
,
> and z
is some set in [
+
.
Then 1
P( z)
No(
,
R
) = .
Proof. Let
T
z
be a P( z)name for the set of all functions from into 2, i.e.,
it satises that 1
P( z)
T
z
=
2. We prove that
(A) 1
P( )
for every f
2 there is g
T
z
with f
,
R
g.
This suces since then
1
P( )
card(
T
z
/
,
R
) card(
2/
,
R
) = No(
,
R
) = ,
and by Lemma 4.2, we can conclude
1
P( z)
No(
,
R
) = card(
T
z
/
,
R
) = .
Now assume, contrary to (A), that (4.1) on page 12 holds and there are a
condition p in P( ) together with a P( )name for a function from into
2 such that
(B) p
P( )
for all g
T
z
, ,
,
R
g .
Without loss of generality, the name has of cardinality . By Lemma 4.2,
and since each cardinal in is listed times, we may choose p and the name
so that the coordinates appearing in adds a tree with the same number
of branches as some coordinate in dom( z) does, i.e., for every
1st
,
there is dom( z) with
so that 1
P( )
T
z
. Hence, by applying (B),
we get
p
P( )
,
,
.
(ii) We dene P( )names
 <
+
) for functions from into 2, and
conditions q
 <
+
) in P( ).
(iii) For every <
<
+
we dene a mapping
,
determines an automorphism
,
(
R) =
R,
,
(p) = q
,
,
() =
, and
,
) =
P( )
,
,
.
16
7
1
9
r
e
v
i
s
i
o
n
:
2
0
0
2

0
1

1
2
m
o
d
i
f
i
e
d
:
2
0
0
2

0
1

1
2
(iv) Finally, we x a P( )generic set G over V and, by applying a stan
dard density argument, we show that for some B [
+
]
+
, all the
conditions q
R
1st

>
=
R
1st

> ;
R
1st
.
Remark. Of course we would like to have that q
=
,
<
+
. Unfortunately, that is not possible since it might be the case
that for some
>
,
, z
) < ).
(i) We dene the name
;
ran()
z
;
for every dom(
1st
),
(
1st
())
=
R
is identity (implying
1st
(
>
) is identity);
for every dom(
1st
)
is identity;
for every dom(
1st
)
>
and dom(
() ,
p
;
for every dom(
1st
)
,
1st
() , dom(p)
R
1st
1st
and
is
some injective function having range z
.
It is possible to fulll these conditions by the choice of , because of the
cardinality demands on z, and since
p
2. However,
is a
P( z)name, so 1
P( )
T
z
holds, too.
(ii) For every <
+
, we dene a mapping
is
is
)
z
, when <
+
, it is possible to choose
) =
R
is identity;
for every dom(
1st
),
(
1st
())
=
;
for every dom(
1st
) (
),
is identity;
for all dom(
1st
)
>
, the sets (
) and ran(
R
1st
1st
1st
dom(p)) and ran(
1st
)
R
1st
, for all
<
+
, are pairwise disjoint.
(iii) Fix indices <
<
+
. Consider the set of pairs x, y) satisfying that
x dom(
) and
(x) = y, or
there is z dom() =
(z) = y.
Because of the conditions given above, we have that
for all dom(
1st
) = dom(
1st
),
1st
() =
1st
() i
1st
() = ;
for all , ) dom(
) = dom(
),
() =
() i
() = ;
for all ,= dom(
1st
),
() ,=
();
for all , ) ,= , ) dom(
),
() ,=
().
18
7
1
9
r
e
v
i
s
i
o
n
:
2
0
0
2

0
1

1
2
m
o
d
i
f
i
e
d
:
2
0
0
2

0
1

1
2
Hence the set of pairs we considered is the following welldened injective
function from Mps( ):
=
_
(
dom()) ()
1
_
.
We let the mapping
,
Mps( ),
dom(
,
1st
) = , and for each < , dom(
,
) =
. It follows that
,
(
R) = (
R) =
(
R) =
(
R) =
R;
,
(p) =
(p) = q
R
) = );
,
() =
() =
R
) = );
,
) =
1
(
)
_
=
() =
(remember, that () =
).
(iv) Our demands on the mappings
, <
+
, ensure that for each , )
p
, if , )
(q
)
then b
(q
)
,
= b
p
,
. Therefore, p and q
are compatible
conditions. Moreover, for every <
+
, the set
D
is a dense set below the condition p (which means that for every s p there
is r s with r D
). Since p G, D
G must be conal in
+
. So B
has cardinality
+
. 4.3
4.2 Isomorphism classes of names
First of all we x z so that the subforcing P( z) of P( ) satises the assump
tions of Lemma 4.3. Secondly we x P( z)names
,
,
.
Since P( z) has
+
c.c., we may assume that each of the names
has car
dinality .
Denition 4.4 For every < we x an enumeration
i
,
i
)  i < )
of
and
i
=
i
;
for every i < and =
i
=
i
, if
i
=
i
;
for all , )
R
and i < ,
i
,
i
) = , ) i
i
,
i
) = , ).
(
) =
and
i
,
i
) =
i
,
i
) for each i < .
For every < we denote the set < 
by
. Now by the
choice of P( z), and the assumptions
<
= and 2
=
+
, the number of
nonisomorphic names in
 < is
+
, i.e., the cardinality of the
family
 < is at most
+
.
Let be a subset of such that card()
+
and
 is a set
of representatives of the isomorphism classes. Since >
+
, the following
equation holds:
(4.3) =
_
card(
).
Dene the set of all small cardinals to be
SC(
R) =
R
1st
and
.
Note that this set might be empty. Anyway, we know that
(4.4) max
++
, supSC(
R).
To prove that is a cardinal in
we shall show that for every ,
the cardinality of
of so that card(
) has one
of the following form: either card(I
) < and
(4.5) card(
)
_
_
iI
i
_
_
iI
i
_
(Subsection 4.3),
or else, card(I
) = and
(4.6) card(
) =
_
K[I
]
<
card
_
iK
i
_
(Subsection 4.4) .
This suces, since we may ensure that for every and for each i I
,
the cardinal
i
is in SC(
R). This means that only those small cardinals are
20
7
1
9
r
e
v
i
s
i
o
n
:
2
0
0
2

0
1

1
2
m
o
d
i
f
i
e
d
:
2
0
0
2

0
1

1
2
used whose coordinates appear in the name
R. It follows that there occurs
at most dierent cardinals in the union (4.3). Hence, for some sequence
X
k
 k < ) of sets in [SC(
R)]
<
,
=
_
k<
card
_
X
k
_
.
Remark. From our assumption that for every (
++
) and <
, the inequality
+
holds, it follows that is either supSC(
R) or
card
_
X
_
for some subset X of SC(
R) with card(X) < .
4.3 Case 1: The parameter depends on less than coordi
nates
For the rest of the proof, let
, is
greater or equal to the lower bound given in (4.4) on the preceding page,
i.e.,
and card(
) is at least max
++
, supSC(
R).
To simplify our notation, let
i
 i < ) and
i
 i < ) denote
the sequences
and
by
.
Dene the set of all critical indices of the isomorphism class of
to be
(4.7) J
= i < 
i
and
i
,
i
) ,
R
.
Note that for every
, the equations
and
( J
) =
(J
i <  z
i
=
i
i < 
R
1st
. Thus
i
 i J
SC(
R) holds, too.
The set J
such
that
and
are
names for nonequivalent functions ((4.2) on page 19). For a similar reason
card
_
iJ
i
_
card(
) holds.
Now suppose that already some subset K of J
i
_
card(
).
21
7
1
9
r
e
v
i
s
i
o
n
:
2
0
0
2

0
1

1
2
m
o
d
i
f
i
e
d
:
2
0
0
2

0
1

1
2
If card(
) = card(
iK
i
) we can dene I
to be K. Otherwise, our
assumption on the cardinal arithmetic gives
card
_
iK
i
_
=
_
_
iK
i
_
+
> card(
).
By the choice of
, card(
) supSC(
R)
iK
i
. Hence card(
) =
iK
i
and again we can choose I
to be K.
It follows, that when card(J
) < we can nd I
satisfying (4.5) on
page 20.
4.4 Case 2: The parameter depends on coordinates
Remark. If is such that each
is
+
or , we have so far proved that
must be either
+
or = .
For the rest of the proof we assume that the set J
]
<
, card
_
iK
i
_
<
card(
). So
card(
) holds, where
=
_
K[J
]
<
card
_
iK
i
_
.
We prove that card(
,
i
i
,
for each i J
,
i
=
i
, and
for every i < j < ,
i
,
i
) ,=
j
,
j
).
22
7
1
9
r
e
v
i
s
i
o
n
:
2
0
0
2

0
1

1
2
m
o
d
i
f
i
e
d
:
2
0
0
2

0
1

1
2
Again, to simplify our notation, we write (
i
(
i
) 
i < ) when
is in c
i
,
i
)  i <
dom(),
Every sequence in c
determines a P( z)name
for a function from
into 2. Namely, we dene
to be the name (
i
,
i
)  i < and (
) = .
A pair
, ) of sequences in c
i
,
i
) ,=
j
,
j
).
Denote the set i J

i
=
i
, for
, c
, by A(
, ).
The sequence
is in c
when
. Also
is the name
for every
. In fact,
 c
.
Lemma 4.6 Suppose
1
,
2
,
1
,
2
c
1
,
1
) and
2
,
2
) are neat, and moreover, A(
1
,
1
) = A(
2
,
2
) holds. Then there is an
automorphism of P( z) such that (
R) =
R, (
1 ) =
2 and (
1) =
2.
Hence for every p P( z),
p
P( z)
1
,
R
1 i (p)
P( z)
2
,
R
2.
Proof. There is a mapping in Mps( z) such that (
1
) =
2
and (
1
) =
2
,
because the sequences in c
1
,
1
) = A(
2
,
2
) holds. Furthermore, can be
chosen so that
R
is identity and each
i
is a permutation of z
i
. Hence
determines an automorphism as wanted.
For technical reasons we dene
/
=
_
I  there are ,=
such that
) is neat and I A(
)
_
.
The next lemma explains why we closed the set /
.
23
7
1
9
r
e
v
i
s
i
o
n
:
2
0
0
2

0
1

1
2
m
o
d
i
f
i
e
d
:
2
0
0
2

0
1

1
2
Lemma 4.7 For all
, c
, if
, ) is neat and A(
, ) is in /
, then
1
P( z)
,
,
R
.
Proof. First we x ,=
) is neat and I =
A(
, ) A(
). Let
be a sequence in c
I =
I,
i
,
j
 j < . Then the pair
) is neat
and A(
,
,
, because then
it follows from Lemma 4.6 that 1
P( z)
,
,
R
.
Suppose, contrary to this claim, that p P( z) satises
p
P( z)
.
Let J denote the set A(
from c
so that
J =
J,
i
,
p
j
 j <
j
 j < .
Then the pair
) is neat and A(
and
,
,R
. Once more, it
follows from Lemma 4.6 that
1
P( z)
,
,R
.
Choose from Mps( z) so that (
R) =
R, (
) =
, (
) =
,
_
[
p
]
p
_
is identity, and determines an automorphism of P( z). This is
possible by the choice of the sequence
. Since A(
) = A(
) and
R
.
Now there is q P( z) satisfying q p and q (p). Since
,
R
is a name
for an equivalence relation, q
P( z)
R
, a contradiction.
Next we want to show that there is always a small set of indices outside of
/
.
Lemma 4.8 When J
, ) of sequences in c
satisfying that
A(
, ) [J
]
<
and p
P( z)
R
.
24
7
1
9
r
e
v
i
s
i
o
n
:
2
0
0
2

0
1

1
2
m
o
d
i
f
i
e
d
:
2
0
0
2

0
1

1
2
Proof. First of all, for every i J
and
i
2 we x an ordinal
from
i
so that for all i, j J
,
i
2, and
j
2,
i i = j
and = . Fix also a coordinate < so that
> and , Z ( is
outside of P( z)). Suppose G is a P
2)
V[G]
, let
u
denote the following sequence:
u
=
u
i
 i < ),
u
i
=
ui
if i J
, and
u
i
=
i
otherwise. Then each of the sequences
u
is
in (c
)
V[G]
. Moreover,
u
,
v
) is a neat pair of for all u and v in (
2)
V[G]
.
Let H be a P( z)generic set over V[G]. In V[G], there are at least
dierent
functions from into 2. By the assumption (4.1) on page 12 and Lemma 4.2,
there are only equivalence classes of
,R
in V[G][H]. It follows, that for
some p H and u ,= v (
2)
V[G]
the following holds in V[G],
p
P( z)
u
,
v .
By the denition of the ordinals
, we have that A(
u
,
v
) = i J

ui = vi [J
]
<
, and hence A(
u
,
v
) is in V.
Now, in V, we can x a neat pair
1
,
2
) of sequences in c
such that
A(
1
,
2
) = A(
u
,
v
). Let (Mps( z))
V[G]
be such that it determines,
in V[G], an automorphism of P( z) satisfying (
R) =
R, (
u
) =
1
, and
(
v
) =
2
. For such a in V[G], we have that (p)
P( z)
1
,
R
2. Note,
that the condition q = (p) is in V. From the equivalence of the forcings
P
P( z) and P( z) P
R
2.
) =
) >
. In the lemma
below, we show that then all the subsets of J
.
It follows from Lemma 4.7, that for all
, c
, if
, ) is neat and A(
, )
is of cardinality < , then 1
P( z)
,
,
R
. By Lemma 4.8, this leads to
a contradiction. So it remains to prove the following last lemma.
Lemma 4.9 If card(
) >
then [J
]
<
/
.
Proof. Fix a set K from [J
]
<
. Since
card(
) >
card
_
iK
i
_
2
,
25
7
1
9
r
e
v
i
s
i
o
n
:
2
0
0
2

0
1

1
2
m
o
d
i
f
i
e
d
:
2
0
0
2

0
1

1
2
there is X
1
of cardinality (2
)
+
such that for all ,= X
1
, K
A(
)
+
such that for all ,=
X
2
, the intersection
i
 i <
i
 i < is some xed set
. There are also I and X
3
[X
2
]
(2
)
+
such that for all X
3
,
i < 
i
= I. Hence there is ,= X
3
with
I =
I and
i
 i I
i
 i I = , i.e.,
). 4.9
5 Remarks
The following facts are also useful to know, when applying the theorem
proved. Write Fn(, 2, ) for the ordinary Cohenforcing which adds a
generic subset of , i.e., the forcing
 is a partial function from into 2 and card() <
ordered by reverse inclusion.
Fact 5.1
(a) There is a dense subset Q Fn(, 2, ) and a dense embedding of Q
into P
(where P
has cardinality .
(c) The forcing P( ) is locally Cohen, i.e., every subset Q of P( ) of
size is included in a complete subforcing Q
of P( ) so that Q
is
equivalent to Fn(, 2, ).
(d) Assume that is a weakly compact cardinal, and V is such that re
mains weakly compact after forcing with Fn(, 2, ). Then every locally
Cohen forcing preserves weakly compactness of .
Note, that if is a weakly compact cardinal then, using upward Easton
forcing, it is possible to have a generic extension V[H] such that is weakly
compact in V[H] and remains weakly compact in all extensions V[H][G],
where G is Fn(, 2, )generic over V[H] (an unpublished result by Silver).
These facts are applied in [SV].
26
7
1
9
r
e
v
i
s
i
o
n
:
2
0
0
2

0
1

1
2
m
o
d
i
f
i
e
d
:
2
0
0
2

0
1

1
2
References
[Jec71] Thomas J. Jech. Trees. J. Symbolic Logic, 36:114, 1971.
[Jec97] Thomas Jech. Set theory. SpringerVerlag, Berlin, second edition,
1997.
[Shea] Saharon Shelah. On nice equivalence relations on 2
. Preprint, No.
724 in the list of Shelahs publications.
[Sheb] Saharon Shelah. Strong dichotomy of cardinality. Results in Math
ematics, to appear.
[SV] Saharon Shelah and Pauli Vaisanen. The number of L
equivalent
nonisomorphic models for weakly compact. Fund. Math. To
appear. No. 718 in the list of Shelahs publications.
Saharon Shelah:
Institute of Mathematics
The Hebrew University
Jerusalem, Israel
Department of Mathematics
Rutgers University
New Brunswick
NJ, USA
shelah@math.rutgers.edu
Pauli Vaisanen:
Department of Mathematics
University of Helsinki
Finland
pauli.vaisanen@helsinki.fi
27