6

5
9


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







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


Universal graphs at the successor of a singular
cardinal
(revised after proof reading)
Mirna Dzamonja
School of Mathematics
University of East Anglia
Norwich, NR4 7TJ, UK
M.Dzamonja@uea.ac.uk
http://www.mth.uea.ac.uk/people/md.html
and
Saharon Shelah
Institute of Mathematics
Hebrew University of Jerusalem
91904 Givat Ram, Israel
and
Rutgers University
New Brunswick, NJ, USA
shelah@sunset.huji.ac.il
http://www.math.rutgers.edu/shelarch
October 6, 2003
Abstract
The paper is concerned with the existence of a universal graph at
the successor of a strong limit singular of conality
0
. Starting from
the assumption of the existence of a supercompact cardinal, a model
1
6
5
9


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







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


is built in which for some such there are
++
graphs on
+
that
taken jointly are universal for the graphs on
+
, while 2

+
>>
++
.
The paper also addresses the general problem of obtaining a frame-
work for consistency results at the successor of a singular strong limit
starting from the assumption that a supercompact cardinal exists.
The result on the existence of universal graphs is obtained as a specic
application of a more general method.
1
0 Introduction
The question of the existence of a universal graph of a certain cardinality and
with certain properties has been the subject of much research in mathematics
([FuKo], [Kj], [KoSh 492], [Rd], [Sh 175a], [Sh 500]). By universality we
mean here that every other graph of the same size embeds into the universal
graph. In the presence of GCH it follows from the classical results in model
theory ([ChKe]) that such a graph exists at every uncountable cardinality,
and it is well known that the random graph ([Rd]) is universal for countable
graphs (although the situation is not so simple when certain requirements
on the graphs are imposed, see [KoSh 492]). When the assumption of GCH
is dropped, it becomes much harder to construct universal objects, and it is
in fact usually rather easy to obtain negative consistency results by adding
Cohen subsets to the universe (see [KjSh 409] for a discussion of this). For
some classes of graphs there are no universal objects as soon as GCH fails
suciently ([Kj], [Sh 500](2)), while for others there can exist consistently a
small family of the class that acts jointly as a universal object for the class at
the given cardinality ([Sh 457], [DjSh 614]). Much of what is known in the
absence of GCH is known about successors of regular cardinals ([Sh 457],
[DjSh 614]). In [Sh 175a] there is a positive consistency result concerning
1
This publication is denoted [DjSh 659] in Saharon Shelahs list of publications. The
authors thank the United States-Israel Binational Science Foundation for their support,
the NSF for their grant NSF-DMS97-04477 and Charles Morgan for many useful comments,
as well as the anonymous referee for the careful reading of the original manuscript. Mirna
Dzamonja would like to acknowledge that a major part of this research was done while
she was a Van Vleck Visiting Assistant Professor at the University of Wisconsin-Madison.
Keywords: successor of singular, iterated forcing, universal graph.
AMS 2000 Classication: 03E35, 03E55, 03E75.
2
6
5
9


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







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


the existence of a universal graph at the successor of singular where is
not a strong limit. In this paper we address the issue of the existence of
a universal graph at the successor of a singular strong limit and obtain a
positive consistency result regarding the existence of a small family of such
graphs that act jointly as universal for the graphs of the same size.
In addressing this specic problem, the paper also oers a step towards
the solution of a more general problem of doing iterated forcing in connection
with the successor of a singular. This is the case because the result about
universal graphs is obtained as an application of a more general method.
The method relies on an iteration of (< )-directed-closed
+
-cc forcing,
followed by the Prikry forcing for a normal ultralter T built by the iteration.
The cardinal here is supercompact in the ground model. The idea is that
the Prikry forcing for T can be controlled by the iteration, as T is being built
in the process as the union of an increasing sequence of normal lters that
appear during the iteration. Apart from building T, the iteration also takes
care of the particular application it is aimed at by predicting the T-names
of the relevant objects and taking care of them (in our application, these
objects are graphs on
+
). The iteration is followed by the Prikry forcing for
T, so changing the conality of to
0
. Before doing the iteration we prepare
by rendering its supercompactness indestructible by (< )-directed-closed
forcing through the use of Lavers diamond ([La]). Unlike the most common
use of the indestructibility of where one uses the fact that is indestructible
without necessarily refering back to how this indestructibility is obtained, we
must use Lavers diamond itself for the denition of the iteration. We note
that the result has an unusual feature in which the iteration is not constructed
directly, but the existence of such an iteration is proved and used.
Some of the ideas connected to the forcing scheme discussed in this paper
were pursued by A. Mekler and S. Shelah in [MkSh 274], and by M. Gitik
and S. Shelah in [GiSh 597], both in turn relaying on M. Magidors inde-
pendence proof for SCH at

[Ma 1], [Ma 2] and Lavers indestructibility

method, [La]. In [MkSh 274]3 the idea of guessing Prikry names of an ob-
ject after the nal collapse is present, while [GiSh 597] considers densities of
box topologies, and for the particular forcing used there presents a scheme
similar to the one we use (although the iteration is dierent). The latter
3
6
5
9


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







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


paper also reduced the strength of a large cardinal needed for the iteration
to a hyper-measurable. The dierence between [GiSh 597] and our results
is that the individual forcing used in [GiSh 597] is basically Cohen forcing,
while our interest here is to give a general axiomatic framework under which
the scheme can be applied for many types of forcing notions.
The investigation of the consistent existence of universal objects also has
relevance in model theory. The idea here is to classify theories in model
theory by the size of their universality spectrum, and much research has
been done to conrm that this classication is interesting from the model-
theoretic point of view ([GrSh 174], [KjSh 409], [Sh 500], [DjSh 614]). The
results here sound a word of caution to this programme. Our construction
builds
++
graphs on
+
that are universal for the graphs on
+
, while
2

+
>>
++
and is a strong limit singular of conality
0
. In this model we
naturally obtain club guessing on S

0
for order type , and this will prevent
the prototype of a stable unsuperstable theory Th(

, E
n
)
n<
from having
a small universal family, see [Sh 457], [KjSh 447]. Hence the universality
spectrum at such
+
classies the prototype of a simple unstable theory (the
theory of a random graph), as less complicated than the prototype of a stable
unsuperstable theory, contrary to the expectation. A possible conclusion is
that in order to obtain a classication into as few as possible nicely dened
classes one should concentrate the investigation of the universality spectrum
as a dividing line for unstable theories only on the case
+
with =
<
,
as the case of the successor of a singular is too sensitive to the set theory
involved. Perhaps just working with
+
where =
|T|
is a reasonable (as
this rules out this particular example), or simply admitting the possibility of
Th(

, E
n
)
n<
and the theory of the random graph being incompatible is a
possible approach.
There are several further questions that this paper brings to mind. From
the point of view of model theory it would be interesting to determine which
other rst order theories t the scheme of this paper and from the point of
view of graph theory one would like to improve the result on the existence
of
++
jointly universal graphs to having just one universal graph. Set-
theoretically, we would like to be able to replace an unspecied singular
strong limit by =

, as well as to investigate singulars of dierent conal-

4
6
5
9


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







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


ity than
0
. We did not concentrate here on obtaining the right consistency
strength for our results, suggesting another question that may be addressed
in the future work.
The paper is organised as follows. The major issue is to dene the iter-
ation used in the second step of the above scheme, which is done in certain
generality in 1. We give there a sucient condition for a one step forcing
to t the general scheme, so obtaining an axiomatic version of the method.
In 2 we give the application to the existence of
++
universal graphs of size

+
for the successor of a strong limit singular of conality
0
.
Most of our notation is entirely standard, with the possible exception of
Notation 0.1. For and ordinal and a regular cardinal < , we let
S

def
= < : cf() = .
1 The general framework for forcing
Denition 1.1. Suppose that is a strongly inaccessible cardinal >
0
. A
function h : H() is called Lavers diamond on i for every x and ,
there is an elementary embedding j : V M with
(1) crit(j) = and j() > ,
(2)

M M,
(3) (j(h))() = x.
Theorem 1.2. Laver ([La]) Suppose that is a supercompact cardinal.
Then there is a Lavers diamond on .
Hypothesis 1.3. We work in a universe V that satises
(1) is a supercompact cardinal, = cf()
+
and GCH holds at and
above ,
(2)

= & =
+
and
(3) h : H() is a Lavers diamond.
5
6
5
9


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







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


Remark 1.4. (1) It is well known that the consistency of the above hypoth-
esis follows from the consistency of the existence of a supercompact cardinal.
We in fact only use the -supercompactness of .
(2) With minor changes, one may replace =
+
by being strongly inac-
cessible > .
Denition 1.5. Laver ([La]) We dene

R = R
+

, R

: , < ),
an iteration done with Easton supports, and a strictly increasing sequence

: < ) of cardinals, where R

and

are dened by induction on

< as follows.
If
(1) h() = (P

, ), where and are cardinals and P

is a R
+

-name of
(< )-directed-closed forcing, and
(2) ( < ) [

< ],
we let R

def
= P

and

def
= . Otherwise, let R

def
= and

def
= sup
<

.
The extension in R
+

is dened by letting
p q [Dom(p) Dom(q) & (i Dom(q))(q i p(i) q(i))],
(where p denotes the weaker condition).
Remark 1.6. The forcing R
+

used in this section is Lavers forcing from

[La] which makes the supercompactness of indestructible under any (< )-
directed-closed forcing.
Convention 1.7. Denitions 1.9 and 1.12, Claim 1.13 and Observation 1.14
take place in V
1
def
= V
R
+

.
Observation 1.8.
+
cf() = < =
+
, 2

=
+
for and

= still hold in V
1
, as Rang(h) H(), and is still supercompact.
6
6
5
9


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







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


Denition 1.9. By induction on i

< we dene the family /

i

as the
family of all sequences

Q = P
i
, Q

i
, A

i
: i < i

= lg(

Q))
which satisfy the following inductive denition, and we let /

def
=

i<
= /
i

.
(1) P
i
H() (and each P
i
is a forcing notion, which will follow from the
rest of the denition),
(2) each P
i
satises the -cc and for i j the forcing notion P
i
is completely
embedded into P
j
by the identity function,
(3) Q

i
is a P
i
-name of a forcing notion (hence a partial order with the least
element
Q

i
) which is a member of H

() (hence of cardinality [[),

(4) If cf(i) , then P
i
=

j<i
P
j
,
(5) A

i
is a canonical P
i+1
-name of a subset of ,
(6) Letting G
i
be P
i
-generic over V
1
, then in V
1
[G
i
],
(a) we let NUF
def
= T : T a normal ultralter on , and
(b) for every T NUF we are given a (< )-directed-closed forcing
notion Q
i
D
H()
V
1
[G
i
]
whose minimal element is denoted by
Q
i
D
,
(7) With the notation of (6), we have that Q

i
[G
i
] is
NUF
_
T Q
i
D
: T NUF
_
,
(8) The order on Q

i
[G
i
] is given by letting
x y i [x = y or x = or (x = T NUF & y x Q
i
D
) or
x = (T, x

), y = (T, y

) for some T NUF and Q

i
D
[= x

],
7
6
5
9


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







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


(9) We have (we adopt the usual meaning of canonical below, see [Sh -f],
I. 5.12. for the exact denition)
P
i
def
=
_

_
p :
(i) p is a function with domain i
(ii) j Dom(p) = p(j) is a canonical P
j
-name
of a member of Q

j
(iii) [SDom(p)[ < (see below)
_

_
,
ordered by letting
p q [Dom(p) Dom(q) & (i Dom(q))(q i p(i) q(i))],
where
(Denition 1.9 continues below)
Notation 1.10. (A) For i < i

, and p P
i
, we let the essential domain of
p be
SDom(p)
def
=
_
j Dom(p) :
_
p j
P
j
p(j) NUF

(T

,
Q
j
D

) : T

NUF

__
,
(B) For i < i

and p P
i
we call p purely full in P
i
i:
SDom(p) = and for every j < i we have
p j
P
j
p(j) NUF

.
If i is clear from the context we may say that p is purely full in its
domain.
(C) Suppose that i < i

and p P
i
is purely full in P
i
, we dene
P
i
/p
def
=
_
q P
i
:
p q & for each j < i,
q j q(j) is of the form (T

, x) for some x
_
,
with the order inherited from P
i
.
(Denition 1.9 continues:)
8
6
5
9


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







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


(10) For every i i

and p P
i
which is purely full in P
i
we have that P
i
/p
satises -cc and P
i
/p H().
Observation 1.11. If

Q /

and i < lg(

Q), then P
i+1
= P
i
Q

i
.
Denition 1.12. (1) We dene the family /
+

as the family of all sequences

Q = P
i
, Q

i
, A

i
: i < )
such that
i < =

Q i /

.
We let P

def
=

i<
P
i
.
(2) Suppose that

Q /
+

and p
i
: i < ) with p
i
P

i
are purely full in
P

i
and increasing in P

, where
i
def
= min : p
i
P

(so if i < j then

p
i
= p
j

i
). We dene
P

/
i<
p
i
def
= q P

: (i < )[q
i
P

i
/p
i
] ,
with the order inherited from P

.
Claim 1.13. (1) If

Q /

, then for all i lg(

Q), we have that P
i
is
(< )-directed-closed.
(2) Similarly for

Q /
+

.
Proof of the Claim. (1) Given a directed family p

: <

< of
conditions in P
i
, we shall dene a common extension p of this family. Let
us rst let Dom(p)
def
=

<

Dom(p

). For j Dom(p), we dene p(j) by

induction on j. We work in V
P
j
1
and assume that p

j : <

G
P
j
.
If j /

<

SDom(p

), then notice that there is at most one T ,= such

that for some (possibly more than one) <

we have p

j p

(j) = T

,
as the family is directed. If there is such T, we let p(j)
def
= T, otherwise we
let p(j) = .
If j

<

SDom(p

), similarly to the last paragraph, we conclude that

there is a name T

such that
[ <

& j SDom(p

)] = p

j p

(j) T

j
D
.
9
6
5
9


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







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


As Q

j
D
is forced to be (< )-directed-closed (see (6)(b) of Denition 1.9), we
can nd in V
P
j
1
a condition q such that q p
j
() for all <

such that
j SDom(p

). Let p(j)
def
= (T

, q) for some such q.

(2) Follows from (1) as = cf() > .
1.13
Observation 1.14. Suppose that

Q /
+

, i < j < and p P

i
, q P
j
are purely full in their respective domains, while p q. Then
(1) Dom(p) = i j = Dom(q) and Dom(p) = p() = q().
(2) Suppose that r P
i
/p.
Then dening r +q P
j
by letting Dom(r +q) = Dom(q) and letting
for Dom(r)
(r +q)()
def
=
_
r() if Dom(p)
q() otherwise,
we obtain a condition in P
j
/q.
(3) For r
1
, r
2
P
i
/p we have that
() r
1
and r
2
are incompatible in P
i
/p i r
1
+q and r
2
+q are incom-
patible in P
j
/q,
() r
1

P
i
/p
r
2
r
1
+q
P
j
/q
r
2
+q.
(4) P
i
/p <
f
P
j
/q where f(r)
def
= r +q. We also write f = f
p,q
.
(5) Suppose that the sequence p = p
i
: i < ) satises that each p
i
P

i
is purely full in P

i
, and the sequence p is increasing in P

, where

i
def
= min : p
i
P

and
i
: i < ) is strictly increasing. Then P

= P

/
i<
p
i
is
isomorphic to the limit of a (< )-supported iteration of (< )-directed-
closed -cc forcing.
10
6
5
9


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







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


(6) For every r P

, there is q r with SDom(q) = SDom(r) and p purely

full in some P
i
, such that q P
i
/p.
Convention 1.15. This Convention applies to Observation 1.14(5) above.
(a) Justied by Observation 1.14(5), in the case that each
i
=
i
+ 1 in the
sequel we may abuse the notation and write
P

limP

i
/(p
i

i
), Q

i
p
i
(
i
)
: i < ),
even though this is not literally an iteration of forcing (since the iterands
are not specied at each coordinate). We do this to emphasize the sequence
Q

i
p
i
(
i
)
: i < ), whose importance will become clear in the Main Claim 1.18.
(b) Since f
p,q
are usually clear form the context we simplify the notation by
not mentioning these functions explicitly.
Claim 1.16. Suppose that

Q /
+

and t

is a P

-name of an ordinal, while

p P

is purely full in its domain.

Then for some j < and q we have p q P
j
, and q is purely full in
P
j
, and above q we have that t

is a P
j
-name (i.e t

is a P
j
/q-name).
Proof of the Claim. Given p P

purely full in its domain, and suppose

that the conclusion fails. Let i < be such that p P
i
. We shall choose by
induction on < ordinals i

and

and conditions p

and r

such that
(i) i

[i, ) and i

: < ) is increasing continuous,

(ii) p

P
i

is purely full in P
i

, with p
0
= p and p

for ,
(iii) p

with r

P
t

,
(iv) r

is incompatible with every r

for < ,
(v) p

def
=
<
p

for a limit,
(vi) r

P
i
+1
/p
+1
.
11
6
5
9


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







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


We now explain how to do this induction.
Given p

and i

. Since we are assuming that t

is not a P
i

-name above
p

, it must be possible to nd r

and

as required. Having chosen r

,
(by extending r

if necessary), we can choose p

+1
as required in item (vi)
above, see Observation 1.14(6). This determines i
+1
. Note that i
+1
< as
P

def
=

j<
P
j
.
However, completing the induction we arrive at a contradiction, as let-
ting p

def
=
<
p

we obtain a condition purely full in its domain. Hence

P
def
= P
sup
<
i

/p

has -cc, but r

+ p

: < forms a set of pairwise

incompatible conditions in P.
1.16
Convention 1.17. Now we go back to V , i.e. the Main Claim 1.18 takes
place in V .
Main Claim 1.18. Suppose
()

Q

= P

i
, Q

i
, A

i
: i < ) is an R
+

-name for a member of /

,
() j : V M is an elementary embedding such that

M M, crit(j) = ,
< j() and
(j(h))() = (P

, )
(such a choice is possible by the denition of Lavers diamond) .
Considering j(R
+

, R

: < )) in M (as for < we know that

R
+

, R

: < ) H()), by its denition we see that

j(R
+

, R

: < )) = R
+

, R

: < j())
and R

= P

while

= . Hence j(R
+

) = R
+

for some R
+

-
name R

M for a forcing notion, which is forced to be (< )-directed-

closed.
We also let

= P

i
, Q

i
, A

i
: i < j())
def
= j(P

i
, Q

i
, A

i
: i < )).
Then in V
R
+

, the following holds: we can nd =

i
: i < ),
p

= p

i
: i < ) and q

= q

i
= (
1
q
i
,
2
q
i
) : i < ) such that
12
6
5
9


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







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


(a)
i
: i < ) is strictly increasing continuous and each
i
< ,
(b) p

i
P

i
+1
is purely full in P

i
+1
,
(c) p

is increasing in P

,
(d) For every i < , we have q

i M
R
+

, and in M
R
+

we have
(p

i
,
1
q
i
,
2
q
i
) P

j(
i
+1)
,
while (p

i
,
1
q
i
) P

,
(e) In M
R
+

we have that for <

(p

i
,
1
q
i
,
2
q
i
) : i < ) is increasing in P

sup
i<
j(
i
+1)
,
(f) In M
R
+

, it is forced by (p

i+1
,
1
q
i+1
) that
2
q
i+1
is an upper bound to
j(r) : r G

P
i
Q

i
+1
p

i
(
i
)
,
(g) If B

is an R
+

-name of a P

i
+1
-name of a subset of , then for some
R
+

-name t

for a truth value (i.e. an ordinal 0, 1, 1 standing

for true and 0 for false):
(1) In V we have that (
R
+

, p

i+1
) forces t

to be a P

i+1
+1
/p

i+1
-name,
(2) M [= [(
R
+

, p

i+1
, q

i+1
) j(B

) i t

= 1].
(h) In M
R
+

, either
(p

i+1
, q

i+1
) j(A

i
),
or p

i

P
that
there is no q = (
1
q,
2
q)
R

j(
i
)+1
q

i
with
1
q
R

2
q(j(
i
))
P

j(
i
)+1
j(r) : r G

P
i
Q

i
p

i
(
i
)

and j(A

i
).
[Note that j(A

i
) is a P

j(
i
)+1
-name for a subset of j().]
13
6
5
9


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







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


(i) If cf(i) , then in V
R
+

i
we have p

i
(
i
) NUF and specically
p

i
(
i
) =
_
B

[G
P
i
] :
B

is a P

i
/(p

i

i
)-name for a subset of
and t

[G
P
i
] = 1
_
.
Proof of the Main Claim. Consider R
+
i
, R

i
,
i
: < i < j()) over
R
+

in M. By the inductive denition of R
i
(see Denition 1.5), which
is preserved by j, we have that R

i
is a name for the trivial forcing whenever
i is not such that ( < i)

< i. Since (j(h))() = (P

, ) we have that
for every i satisfying < i < , R

i
is a name for the trivial forcing. For
i < j(), we have that R

i
is a name for a (< )-directed-closed forcing
in M, so in V as well, as
<
M M. Similarly we conclude that P

j()
names a (< )-directed-closed forcing notion, for all < . This observation
will be used repeatedly and in particular will enable us to use the master
condition idea in the induction below. In particular, we can conclude that
R

is (< )-directed-closed. By the choice of j,

j(R
+

)
each P

i
/p is (< )-directed-closed for p P

i
purely full in P

i
.
Now we choose (
i
, p

i
, q

i
) M
R
+

by an induction on i carried in V . We
start with
0
= 0, p

0
P
1
any condition purely full in P
1
, and q

0
= .
Choice of p

i+1
, q

i+1
and
i+1
.
Given p

i
and
i
in V
R
+

. We have (recall Convention 1.15) that

p

i

i

P
i
[Q

i
p

i
(
i
)
[ & G

i
p

i
(
i
)
P

i
+1
/p

i
.
Hence in M, letting X

i
def
= j(r) : r G

P
i
Q

i
p

i
(
i
)
we have
(
R
+
j()
, j(p

i

i
))
P

j(
i
)
X

i
P

j(
i
)+1
is directed and j(p

i
(
i
)) [X

i
[ .
In V
1
, we have that the forcing P

i
+1
/p

i
is a -cc forcing notion of size ,
hence there are

= canonical P

i
+1
/p

i
-names for a subset of .
Let us enumerate them in a limit order type as B

i+1

: <

(i + 1) ),
with
B

i+1
0
= A

i
. ()
14
6
5
9


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







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


This choice for i + 1 helps us to fulll clause (h) for i. By induction on
<

(i + 1) we choose p
i+1

purely full in its domain, increasing continuous

with , q
i+1

= (
1
q
i+1

,
2
q
i+1

) increasing with ,
i+1

increasing with and

t

i+1

as follows.
Let p
i+1
0
def
= p

i
,
i+1
0
def
=
i
and q
i+1
0
def
= q

i
.
Coming to +1, let Gbe a R
+

generic over V such that (
R
+

, p
i+1

) G.
In M[G] we ask the -question:
Is it true that there is no q satisfying the following condition ()
q
, which
means
() q = (
1
q,
2
q)
R

j(
i+1

)+1
q
i+1

: and
()
1
q
R

2
q X

i
& j(B

i+1

) &
2
q P

j(
i+1

)+1
/(j(p
i+1

) j(
i+1

)+1)?
(Here
2
q X

i
means that
2
q is above every condition in X

i
, which then
guarantees that clause (f) is satised).
Case 1. If the answer is positive, i.e. for no q do we have that ()
q
holds
in M, we dene t

i+1

def
= 0 (hence a R
+

-name for a truth value), and
dene
q
i+1
+1
= (
1
q
i+1
+1
,
2
q
i+1
+1
)
to be any R
+

-name for a condition in R

j()
such that
(
R
+

, p
i+1

)
1
q
i+1
+1

R

1
q
i+1

for every , and
(
R
+

, p
i+1

,
1
q
i+1
+1
)
P

j(
i+1

)+1

2
q
i+1
+1

2
q
i+1

: &
2
q
i+1
+1
X

i
.
The choice of
1
q
i+1
+1
is possible by the induction hypothesis and the fact that

R
+

is (< )-directed-closed.
Let us verify that the choice of
2
q
i+1
+1
is possible. Working in M we have that
(
R
+

, p
i+1

,
1
q
i+1
+1
) forces X

i
to be a (< )-directed subset of P

j()
of size < .
Hence if = 0 we can choose
2
q
i+1
to be forced to be above X

i
. We can
similarly choose
2
q
i+1
+1
for > 0.
15
6
5
9


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







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


Case 2. If the answer to the question is negative, so there is q satisfying
()
q
, we let t

i+1

def
= 1 and choose q
i+1
+1
= (
1
q
i+1
+1
,
2
q
i+1
+1
) in M exemplifying
the negative answer.
At any rate, t

i+1

is a R
+

-name for an ordinal. By Claim 2.19, in
V
R
+

there is
i+1
+1

i+1

and a purely full in its domain p

i+1
+1
p
i+1

with
p
i+1
+1
P

i+1
+1
such that t

i+1

is a P

i+1
+1
/p
i+1
+1
-name.
For limit, let
i+1

def
= sup
<

i+1

, p
i+1

def
=
<
p
i+1

, and q
i+1

not dened.
At the end, we let
i+1
def
= sup
<

(i+1)

i+1
+1
and p

i+1
any purely full
condition in P

i+1
+1
with p

i+1

<

(i+1)
p
i+1

, and q

i+1
such that
(
R
+

, p

i+1
) q

i+1

R

j(
i+1
)+1
q
i+1

: <

(i + 1).
Choice of p

i
, q

i
and
i
for i < limit. We let
i
def
= sup
j<i

j
and choose
p

i
P

i
+1
purely full so that p

i

ji
p

j
, and if cf(i) , then
p

i
(
i
)
def
=
_
B

[G
P
i
] : B

is a P

i
/(p

i

i
)-name for a subset of
and t

[G
P
i
] = 1
_
.
Recall that
>
M M so (
j
, p

j
, q

j
) : j < i) M. Condition (f) is
satised by the denition of the order in P

i
(and j(P

i
)). It follows by
the construction and standard arguments about elementary embeddings and
master conditions that
p

i

i

P
i
p

i
(
i
) NUF

.
Then we can choose q

i
so that (
R
+

, p

i
, q

i
) (
R
+

, p

j
, q

j
) for all j < i and
q

i
j(r) : r G
P
i
, which is again possible by the observation at the
beginning of the proof.
1.18
Conclusion 1.19. In V
1
, if

Q /
+

, p

i
: i < ) and
i
: i < ) are as
guaranteed by Main Claim 1.18, letting T

i
def
= p

i
(
i
), it follows by Observa-
tion 1.14(5) that

P = P

i
/(p

i

i
), Q

i
D

i
: i < )
16
6
5
9


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







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


is an iteration (see Convention 1.15(a)) with (< )-supports of (< )-directed-
closed -cc forcing. In addition, there is a club C of with the property that
in V
P
1
T
i
: i C & cf(i) )
is an increasing sequence of normal lters over , with
[i C & cf(i) ] = P

i
/(p

i

i
) T

i
is an ultralter over .
If < satises cf() > then
i<
p

i
forces over P

that

i<
T

i
us an
ultralter over which is generated by cf() sets.
Denition 1.20. (In V
R
+

) Given

Q = P
i
, Q

i
, A

i
: i < ) /
+

.
We say that

Q is tted i there is a continuous increasing sequence

i
: i < ) of ordinals < , and a sequence p

i
: i < ) of conditions
each purely full in its domain with p

i
P

i
+1
, such that letting T

i
def
= p

i(
i
)
,
P

i
+1
/(p

i

i
), Q

i
D

i
: i < )
is an iteration with (< )-supports of (< )-directed-closed -cc forcing, and
cf(i) =
P

i
+1
/(p

i

i
)
A

i
T

i
.
Crucial Claim 1.21. (In V
R
+

) The following is a sucient condition for

/
+

to be tted:
There is a pair (R, h) such that:
(1) R is a function such that for every forcing P with [P[ in H() and
a P-name T

of a normal ultralter on we have that R[P, T

] is well
dened and is a P-name of a forcing notion of cardinality ,
(2) for every purely full in its domain p P

and i Dom(p), we have that

p i Q

i
p(i)
= R[P
i
/(p i), p(i)],
(3) h is a function such that for every forcing P with [P[ in H()
satisfying the -cc and a P-name T

of a normal ultralter on , h(P, T

)
is well dened and is a P-name h

[P,D

]
of a function h

[P,D

]
: R[P, T

] T

17
6
5
9


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







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


such that for every purely full in its domain (if this makes sense for P)
p P

and i Dom(p) it is forced by p i that:

for every inaccessible

< and every (<

)-directed family g of
conditions in R[P
i
/(p i), p(i)] of size < , such that
r g =

[P
i
/(pi),p(i)]
(r),
there is q g such that q

i
.
Remark 1.22. The condition in Claim 1.21 is sucient for the present ap-
plication in 2. It may be weakened if needed for some future application.
Really, the condition to use instead of it is that in item (h) of Main Claim
1.18, for all i of conality < , we are in the good case, i.e. the rst case
of item (h). However, we wish to have a criterion which can be used without
the knowledge of the proof of the Main Claim 1.18, and the condition in
Claim 1.21 is one such criterion.
Proof of the Crucial Claim. By Conclusion 1.19 it suces to show that
under the assumptions of this Claim, in the proof of Main Claim 1.18 we can
choose
i
: i < ), p

i
: i < ) and q

i
: i < ) so that for every i with
cf(i) , the answer to the 0th question in the choice of q
i+1
1
is negative,
i.e. there is q such that ()
q
holds. The proof is by induction on such i. We
use the notation of Main Claim 1.18.
Given i with cf(i) . Hence we have
p

i
(
i
) =
_
B

[G
i
] :
B

a P

i
/(p

i

i
)-name
for a subset of and t

= 1
_
def
= T

i
.
In M we have
(
R
+

, p

i
, q

i
)
_
j(r)(j(
i
)) : j(r) X

i
is (< ) directed of size <
j(), is inaccessible and (r)[ j(h

[P/(p

i
),p

i
(
i
)]
(r))]
_
.
(The last statement is true by the denition of T

i
and t

, no matter what
h

[P/p

i
,p

i
(
i
)]
(r) is forced to be.)
By the assumption (3) and elementarity, applying j we have that the
answer to the 0th question is negative.
1.21
18
6
5
9


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







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


Denition 1.23. (In V
R
+

) Given = cf() (, ). We dene /

in the
same way as /
+

, but with a freedom of choice for Q

0
. Namely, to obtain the
denition of /

from that of /
+

, we
(A) In item (6) of Denition 1.9, require i > 0,
(B) We let Q
0
be any (< )-directed-closed cardinal preserving forcing
notion in H() that also preserves

= .
Claim 1.24. (In V
R
+

) Main Claim 1.18, Conclusion 1.19, Denition 1.20

and Claim 1.21 hold with /
+

replaced by /

.
Proof of the Claim. As in V
R
+

0
, is still indestructibly supercompact
and

= .
1.24
Discussion 1.25. (1) In the present application, we need to make sure
that cardinals are not collapsed, so we have =
+
and Q
D
is chosen
to have a strong version of
+
-cc which is preserved by iterations with
(< )-supports.
(2) Clearly, Claim 1.21 remains true if we replace the word inaccessible
by e.g strongly inaccessible, weakly compact, measurable.
(3) As we shall see in section 2, the point of dealing with a tted member
of /
+

is to be able to control the Prikry names in the forcing that will

be performed after the iteration extracted from /
+

, namely the Prikry

forcing over
i<
T
i
for some . The point of A

i
is to give us a control
of this ultralter in the appropriate universe.
2 Universal graphs
Theorem 2.1. Assume that it is consistent that there is a supercompact
cardinal.
Then it is consistent to have a singular strong limit cardinal of conality
with 2

+
>
++
, on which there are
++
graphs of size
+
which are
universal for the graphs of size
+
.
19
6
5
9


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







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


Proof. We start with a universe V in which , and satisfy Hypothesis
1.3, with =
+
(in particular
+
< =

+
). Let R
+

be the forcing
described in Denition 1.5. We work in V
R
+

, which we start calling V from

this point on. As we shall not use h and R
+

any more, we free the notation

h and R

to be used with a dierent meaning in this section.
Denition 2.2. Let Q
0
be the Cohen forcing which makes 2

+
= by
adding distinct
+
-branches

: < to (

+
>
2)
V
by conditions of size
, so no cardinal is collapsed and in the resulting universe
each

+
2,
< < =

,=

and
<
+
= [

: < [
+
.
Let =

: <
+
) be xed for the rest of the section, and let us let
V
0
def
= V [G
Q
0
].
Notation 2.3. If T is a normal ultralter on a measurable cardinal , let
Pr(T) denote the Prikry forcing for T.
Discussion 2.4. The idea of the proof is to embed T-named graphs into a
universal graph. We use an iteration of forcing to achieve this. As we intend
to perform a Prikry forcing at the end of iteration, we need to control the
names of graphs that appear after the Prikry forcing, so one worry is that
there would be too many names to take care of by the bookkeeping. Luckily,
we shall not be dealing with all such names, but only with those for which
we are sure that they will actually be used at the end. This is achieved by
building the ultralter that will serve for the Prikry forcing, as the union of
lters that appear during the iteration. To this end, for every relevant T we
also force a set A

that will in some sense be a diagonal intersection of T, so

its membership in the intended ultralter will guarantee that that ultralter
contains T as a subset.
Denition 2.5. Suppose V

V
0
is a universe in which 2

+
= , is xed
as per Denition 2.2, while is measurable and T is a normal ultralter over
. Working in V

, we dene a forcing notion Q
def
= Q
V

D,,
, as follows.
20
6
5
9


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







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


Let

M

= M

=
+
, R

) : < ) list without repetitions all canonical
Pr(T)-names for graphs on
+
. By canonical in this context we mean names
of the form

_
<<
+
/
,


(, )
where each /
,
is a maximal antichain in Pr(T). Then

G
is a subset of
[
+
]
2
and we identify it with a graph g

= g(

) on
+
by letting , form an
edge i < and for some p /
,
G we have (p,

(, ))

or < and
for some p /
,
G we have (p,

(, ))

. (Note that if p Pr(T) and

is
a Pr(T)-name such that p

is a graph on
+
, then there is a canonical
name

as above such that p

.) In the list

M we understand that
M

is the model with universe

+
where R

is the graph relation obtained

by some graph g(

) as above. For deniteness we pick the rst such list in

the canonical well-order of H(). Elements of Q are of the form
p = A
p
, B
p
, u
p
,

f
p
= f
p

: u
p
)),
where
(i) A
p
[]
<
,
(ii) B
p
T T([ (Sup(A
p
))]),
(iii) u
p
[]
<
,
(iv) For u
p
, we have that f
p

is a partial one-to-one function from

+
with
[Dom(f
p

)[ < , mapping Dom(f

p

) to an element of

,
(v) For , u
p
, for every x

, x

, if
f
p

(x

) = f
p

(x

) ,= f
p

(x

) = f
p

(x

),
then for every w [A
p
]
<
0
w, B
p
)
Pr(D)
M

[= R

(x

, x

) i M

[= R

(x

, x

).
In addition, for every w [A
p
]
<
0
and every u
p
and x

, x

Dom(f

p
),
the condition w, B
p
) decides in the Prikry forcing for T if M

satises
R

(x

, x

).
21
6
5
9


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







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


We dene the order on Q by letting p q (here q is the stronger condition)
i
(a) A
p
is an initial segment of A
q
,
(b) A
q
A
p
B
p
,
(c) B
p
B
q
,
(d) u
p
u
q
,
(e) For u
p
, we have f
p

f
q

.
Claim 2.6. Suppose that Q = Q
V

D,,
is dened as in Denition 2.5. Then
in V

:
(1) Q is a separative partial order.
(2) Suppose that G is Q-generic over V

, and let in V

[G]
A

def
=
_
A : (B, u,

f)[A, B, u,

f) G].
Then A

is an unbounded subset of and A

B for every B T.
(3) For < and a
+
, the set
/
a,
def
= p Q : u
p
& a Dom(f
p

)
is dense open in Q.
Proof of the Claim. (1) Routine checking.
(2) For < , the set
1

def
= p Q : ( )[ A
p
]
is dense open in V

, hence A

is an unbounded subset of in V

[G].
For B T the set

B
def
= p Q : B
p
B
is dense open. If p
B
G, then for any q G with q p we have
A
q
A
p
B
p
. Hence A

B A
p
.
22
6
5
9


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







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


(3) Given p Q, clearly there is q p with u
q
. Namely, we may let
for p Q such that / u
p
an extension q be dened by A
q
= A
p
,
u
q
= u
p
, f
q

= and f
q

= f
p

for u
p
. So without loss of
generality u
p
and a / Dom(f
p

). Applying the Prikry Lemma, for

every b Dom(f
p

) and w [A
p
]
<
0
, there is B
w,b
B
p
with B
w,b
T
and such that
(w, B
w,b
)[[
Pr(D)
M

[= bR

a.
Choose < such that (

a, ) /

u
p
Rang(f
p

), which is possible
as for every relevant we have [Dom(f
p

)[ < and [u
p
[ < . Now we de-
ne q by letting A
q
def
= A
p
, B
q
def
=

B
w,b
: w [A]
<
0
& b Dom(f
p

)B
p
,
u
q
def
= u
p
and
f
q

def
=
_
f
p

if ,=
f
p

(a, (

a, )) otherwise.
To verify that q is a condition we discuss 2.5(v). If ,= and
x

, x

Dom(f
q

) then w, B
q
) decides in Pr(T) if M

(x

, x

)
because already w, B
p
) does that. If x

, x

Dom(f
p

) the conclusion
follows similarly. If x

, x

a then the conclusion follows by the

choice of B
q
.
Suppose that f
q

(x

) = f
q

(x

) ,= f
q

(x

) = f
q

(x

) for some x

,= x

and
,= . If x

, x

Dom(f
p

) then
w, B
q
)
Pr(D)
M

(x

, x

) M

because this is already true of w, B

p
). So suppose without loss of
generality that x

= a. But was chosen so that f

q

(a) = (

a, ) is
not in Rang(f
q

), hence the condition 2.5(v) is satised.

2.6
Denition 2.7. (Shelah, [Sh 80]) Let
0
be a cardinal. A forcing notion
P is said to be stationary
+
-cc i for every p

: <
+
) in P, there is a
club C
+
and a regressive h :
+

+
such that for all , C,
[cf() = cf() = & h() = h()] = p

, p

are compatible.
23
6
5
9


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







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


Theorem 2.8. (Shelah, [Sh 80], [Sh 546]) Suppose that
<
=
0
. Iter-
ations with (< )-support of (< )-directed-closed stationary
+
-cc forcing,
are (< )-directed-closed and satisfy stationary
+
-cc.
Claim 2.9. Suppose that Q is as in Claim 2.6. Then Q is (< )-directed-
closed and satises stationary
+
-cc.
Proof of the Claim. First suppose that i

< and p
i
: i < i

is directed.
For i < i

let p
i
def
= A
i
, B
i
, u
i
,

f
i
). We dene A
def
=

i<i

A
i
, B
def
=

i<i

B
i
,
u
def
=
i<i
u
i
, and for u we let f

def
=
i<i
f
i

. It is easily veried that this

denes a common upper bound of all p
i
.
Hence Q is (< )-directed-closed. Now we shall prove that it is
+
-
stationary-cc. Let p
i
: i <
+
) be given, where each p
i
= A
i
, B
i
, u
i
,

f
i
)
and

f
i
= f
i

: u
i
). Let U
def
=

u
i
: i <
+
, hence U and
[U[
+
. Let us x a one-to-one enumeration of U in an order type
+
,
so U =
s
: s < s

+
.
For i <
+
let S
i
= s :
s
u
i
be an increasing enumeration and let

i
def
= otp(S
i
), hence
i
< . For s S
i
let d
i
s
def
= Dom(f
i
s
) and for k <
i
let

i
k
= i =
s
for the k-the element s of S
i
. Let
i
< be given by

i
def
= sup + 1 : ( u
i
)( Dom(f
i

)) (f
i

() = (

, )).
For A []
<
and < dene a language
L
A,
= R
w,k
: w [A]
<
, k < < g
k
: k < P, Q,
where each R
w,k
is a 2-place relation symbol, as is <, each g
k
is a 1-place
function symbol and P, Q are unary predicates. Note that the size of this
language is < .
For i <
+
dene a model N
i
of L
A
i
,
i
with the universe

i
0
_
sS
i
d
i
s
1
and the interpretation given by:
P((a, b)) i b = 0,
24
6
5
9


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







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


Q((a, b)) i b = 1,
< is the partial ordering given by letting (, a) < (, b) i a = b and
< as ordinals,
(, a), (, b) R
w,k
i a = b = 1 and , d
i

i
k
, while
(w, B
i
)
Pr(D)
R

i
k
,
g
k
((, a)) = (, b) i a = 0 = b, = or a = 1, b = 0 and
f
i

i
k
() = (

i
k
, ).
For each relevant A, consider the isomorphism types of models of L
A,
whose universe is a disjoint union of two sets each of size < . There are
such types (because
<
= ), let us enumerate them as
t
A,

: < (A, ) .
For i <
+
let
i
be such that the isomorphism type of N
i
as a model of
L
A
i
,
i
is t
A
i
,
i

i
.
Let F from []
<

>
([
+
]
<
) [
+
]
<
be a bijection onto

+
. Let C be a club of j <
+
such that for j C with cf(j) = we have
F(, A, d
k
: k < ), , , S) < j sup
k<
d
k
, sup(S) < j,
and such that for all i < j we have sups :
s
u
i
< j and
sup
_
Dom(f
i

) : u
i
< j.
Such a club exists because
<
= .
We dene h :
+

+
by letting h(i) = 0 unless i C and cf(i) = ,
when h(i) = F(
i
, A
i
, d
i
s
i : s < i),
i
,
i
, S
i
i). Hence h is regressive.
Suppose that i < j C and cf(i) = cf(j) = are such that h(i) = h(j),
we claim that p
i
and p
j
are compatible. In order to prove this we proceed
with several subclaims.
Subclaim 2.10. A
i
= A
j
and
i
=
j
.
25
6
5
9


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







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


Proof of the Subclaim. This follows from the choice of F and h.
2.10
Let A
def
= A
i
= A
j
,
def
=
i
=
j
.
Subclaim 2.11. N
i
and N
j
are isomorphic as models of L
A,
.
Proof of the Subclaim. This follows from Subclaim 2.10 and the fact that

i
=
j
.
2.11
Subclaim 2.12. If u
i
u
j
then =
i
k
=
j
k
for the same k and =
s
for some s < i.
Proof of the Subclaim. Since u
i
and j C is of conality , we have
that =
s
for some s < j. Hence s < i by the choice of h and so =
i
k
for some k. Since S
i
i = S
j
j we have that =
j
k
as well.
2.12
Subclaim 2.13. If u
i
u
j
and d
i

d
j

, then f
i

() = f
j

().
Proof of the Subclaim. By Subclaim 2.12 there is k such that =
i
k
=
j
k
,
which is
s
for some s < i. By the choice of j we have sup(d
i

) < j, so by the
choice of h we have < i. Since N
i
and N
j
are isomorphic, by the denition
of < in these models we have that (, 1) is a xed point of the isomorphism.
Hence g
k
((, 1)) is as well, so there is a unique such that
f
i

() = (

, ) = f
j

().

2.13
For every w [A]
<
and for every u
i
u
j
and ,

l{i,j},u
l
d
l

we can nd B
,,

such that
w, B
,,

)
Pr(D)
[[M

(,

).
Let B
def
= B
i
B
j

u
i
u
j
,,

d
i

d
j

B
,,

. We claim that a common

upper bound of p
i
and p
j
is given by q = A, B, u = u
i
u
j
,

f) where

f = f

: u) and f

l{i,j}
f
l

. To prove this it suces to prove the

following two claims:
26
6
5
9


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







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


Subclaim 2.14. Suppose that , u and <

are such that

f

() = f

() ,= f

) = f

).
Then for every w [A]
<
we have
w, B)
Pr(D)
M

(,

) M

(,

).
Proof of the Subclaim. We have to do a case analysis.
Case 1. For some l i, j we have that , u
l
and ,

d
l

d
l

.
The conclusion follows by the analogous properties of p
l
.
Case 2. u
i
u
j
, d
i

d
j

and

d
j

d
i

.
We have

/ d
i

, hence

j by the choice of h. Hence

/ d
i

and so

d
j

. In particular
f
j

) = f

) = f

) = f
j

).
Since d
i

d
j

we have [i, j) and so d

i

d
j

and
f
i

() = f

() = f

() = f
i

().
In particular u
i
u
j
. Let

be such that (

, 1) N
i
is the isomorphic
image of (

, 1) under an isomorphism between N

j
and N
i
. Then
f
i

) = f
j

) = f
j

) = f
i

)
and so
f
i

() = f
i

() ,= f
i

) = f
i

).
So for every w [A]
<
we have
w, B
i
)
Pr(D)
M

(,

) M

(,

).
Let w [A]
<
be given. By the choice of the function R
w,k
for k such that
=
i
k
we have
w, B
i
)
Pr(D)
M

(,

) i w, B
j
)
Pr(D)
M

(,

)
and similarly for in place of . Hence
w, B)
Pr(D)
M

(,

) M

(,

)
27
6
5
9


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







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


as required.
Case 3. u
i
u
j
, d
i

d
j

and

d
j

d
i

.
Symmetric to Case 2 with replaced by .
Case 4. u
i
u
j
, d
i

d
j

and

d
j

d
i

.
As in Case 2,

/ d
i

so

j and so

/ d
i

. Hence

d
j

d
j

and
so f

) = f
j

) and f

) = f
j

). Since d
j

we have f

() = f
j

().
Since d
i

we have < j so < i. If d

j

we obtain the desired

conclusion because p
j
Q. But if not, then d
i

, hence u
i
u
j
and so
d
i

i = d
j

j, a contradiction.
Case 5. u
i
u
j
, d
i

d
j

and

d
j

d
i

.
Symmetric to Case 4 with replaced by .
Case 6. u
i
u
j
and

d
i

d
j

while d
j

d
i

.
This case cannot happen because <

.
Case 7. Symmetric to Case 6 with replaced by .
Cannot happen for the same reason as Case 6.
Case 8. u
i
u
j
and ,

d
i

d
j

.
If u
i
u
j
then we are in Case 1. If u
i
then since ,

d
i

we
have ,

< j and so since u

j
we have ,

< i. Hence ,

d
i

and the
conclusion follows because p
i
Q. If / u
i
then u
j
and ,

d
j

d
j

,
hence the conclusion follows as p
j
Q.
Case 9. u
i
u
j
and ,

d
i

d
j

.
Symmetric to Case 8.
Case 10. u
i
u
j
and u
j
u
i
.
Hence ,

d
i

and so ,

< j. Let k be such that =

j
k
and let

=
i
k
. By the choice of h and the fact that N
i
and N
j
are isomorphic, we
have that ,

< i and ,

d
i

, while
w, B
i
)
Pr(D)
M

i w, B
j
)
Pr(D)
M

.
Moreover f
i

() = f
j

() and similarly for

. We get the desired conclusion

by applying this and the fact that p
i
Q.
Case 11. u
j
u
i
and u
i
u
j
.
Symmetric to Case 10.
2.14
Subclaim 2.15. Suppose that u and ,

Dom(f

).
Then w, B)[[
Pr(D)
M

.
28
6
5
9


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







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


Proof of the Subclaim. Follows by the choice of B.
2.15
This nishes the proof of the chain condition.
2.9
Observation 2.16. Suppose that T is a normal ultralter over and Q is
a forcing notion such that

Q
T T

and T

is a normal ultralter over .

Then Pr(T) <
e
Q Pr(T

), where e is the embedding given by

e((a, A))
def
= (
Q
, (a, A)).
Denition 2.17. Suppose that Q is as in Claim 2.6, while Q <P, and
T

is a P-name of a normal ultralter over , extending T A

. For
< we dene Gr

, intended to be a P Pr(T)-name for a graph on

: <
+
(see Claim 2.19 below), dened by letting for
y

, y

: <
+
,
y

i for some p, w, B
p
)) G

with u
p
, p Q and [w] [A
p
]
<
0
and some x

, x

Dom(f
p

)
we have f
p

(x

) = y

and f
p

(x

) = y

,
AND w, B
p
)
Pr(D)
M

[= R

(x

, x

).
Notation 2.18. Suppose that Q is as in Claim 2.6. For < let
f

def
= f
p

: u
p
& p G

Q
.
Claim 2.19. Suppose Q is as in Claim 2.6, while Q <P, and T

is a P-name
of a normal ultralter over , extending T A

(equivalently, A

).
Then
, , A

))
PPr(D

)
f

is an embedding of M

into Gr

.
Proof of the Claim. Let G be P Pr(T

)-generic with , , A

)) G and
suppose that x

, x

are such that M

[= R

(x

, x

) in V [G]. Let p
+
, w, A

))
be a condition in G that forces this. Without loss of generality, we have
p
+
, w, A

)) , , A

)).
29
6
5
9


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







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


In particular, p
+

P
w [A

]
<
0
. Considering P as Q P

/Q, let us write

p
+
, w, A

)) as p, p

, w, A

)). By extending p
+
if necessary, we may assume
that A
p
w, and then using the density of /
x

,
and /
x

,
, we may also
assume that u
p
and x

, x

Dom(f
p

). By extending further, we may

assume that p
+
A

B
p
. Then p
+
, w, A

)) extends p, w, B
p
)), hence
the latter is in G. Since p
P
w, B
p
)[[
Pr(D)
R

(x

, x

), it must be that
w, B
p
)
Pr(D)
M

[= R

(x

, x

). Hence in V [G] we have that

y

= f

(x

)Ry

= f

(x

).
On the other hand, suppose that in V [G] we have y

= f

(x

)Ry

= f

(x

)
and let p, w, B
p
)) exemplify this. In particular, w, B
p
) forces in Pr(T)
that M

[= R

(x

, x

), and since p, w, B
p
)) G, we have that R

(x

, x

)
holds in V [G].
As it is easily seen that each f

is forced to be 1-1 and total, this nishes

the proof.
2.19
Claim 2.20. Suppose that Q and T

are as in Claim 2.19, while G is Q-

generic over V

. Further suppose that H is a Pr(T

)-generic lter over V

[G]
with , A

) H.
Then in V

[G][H], there is a graph Gr

of size
+
such that for every lter
J in Pr(T) satisfying
(
Q
, p) : p J G H

def
= (q, s

) : q G & q s

which is Pr(T)-generic over V

, every graph of size
+
in V

[J] is embedded
into Gr

.
Proof of the Claim. Dene Gr

on
<

: <
+
, hence
[Gr

[ =
+
, by our assumptions on . We let
Gr

[= (

, i)R(

, j) i Gr
D

[= (

, i)R(

, j).
Then Gr

is a well dened graph, as follows by the denition of Q.

Given M a graph on
+
in V

[J]. Let w, A) J force in Pr(T) that M
is a graph on
+
. By Observation 2.16, (
Q
, w, A)) forces in Q Pr(T

) that
30
6
5
9


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







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


M is a graph on
+
, so since (
Q
, w, A)) G H

we have that for some it

is true that M = M

[G][H]. Since (
Q
, , A

)) G H

we have by Claim
2.19 that M embeds into Gr
D

in V

[G][H], but Gr
D

embeds into Gr

by
the denition of Gr

.
2.20
Claim 2.21. Let T be a normal ultralter over and A T. Suppose
that G is a Pr(T)-generic lter over V . Then there is some G

which is
Pr(T)-generic over V and such that (, A) G

while V [G] = V [G

].
Proof of the Claim. Let x = x
G
= s : (B T)(s, B) G, so
G = G
x
= (s, B) Pr(T) : s x
G
s B.
Now we use the Mathias characterization of Prikry forcing, which says that
for an innite subset x of we have that G
x
is Pr(T)-generic over V i
x
G
B is nite for all B T. Hence x A is nite. Let y = x
G
A,
so an innite subset of which clearly satises that y B is nite for all
B T. Let G

= G
y
, so G

is Pr(T)-generic over V and (, A) G

. We
have V [G

] V [G] because y G and V [G] V [G

] because x y is nite.

2.21
Conclusion 2.22. Suppose that Q, T

, G and V

are as in Claim 2.20 and
H is a Pr(T

)-generic lter over V

[G]. Then the conclusion of Claim 2.20
holds in V

[G][H].
Proof. The conclusion follows by Claim 2.20 and Claim 2.21.
2.22
Claim 2.23. Suppose that

Q = P
i
, Q

i
, A

i
: i < ) /

+
is given by
determining Q
0
as in Denition 2.2 and dening Q

i
D

= Q

V [G
P
i
]
D

(as dened in
Denition 2.5) and A

i
= A

i
, where A

i
was dened in Claim 2.6(2).
Then

Q is tted.
31
6
5
9


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







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


Proof of the Claim. We shall take R[P, T

] = Q
V
P
D

,,
if this is well dened
(i.e. V
P
satises the conditions on V

in Denition 2.5) and R[P, T

] =
otherwise. By Claim 1.21, it suces to give a denition of h satisfying the
requirements of that Claim. Suppose that P, T

are such that R[P, T

] is
non-trivial, working in V
P
we dene
h = h
[P,D]
: Q
D
= R
[P,D]
T
by letting h(p)
def
= B
p
for p = (A
p
, B
p
, u
p
,

f
p
). We check that this denition is
as required. So suppose that

< is inaccessible and g is a (<

)-directed
family of conditions in Q
D
with the property that for all p g we have

B
p
. We dene r by letting
A
r
def
=
_
pg
A
p

, B
r
def
=

pg
B
p

, u
r
def
=
pg
u
p
,
and for u
r
, we let f
r

def
=
pg & u
pf
p

. It is easy to check that this

condition is as desired.
2.23
Remark 2.24. The inaccessibility of

was not used in the Proof of Claim

2.23.
Proof of the Theorem nished.
To nish the proof of the Theorem, in V
0
let

Q be as in Claim 2.23. By
Claim 2.23 and the denition of ttedness, we can nd sequences p

i
: i < )
and
i
: i < ) witnessing that

Q is tted. Let T

i
def
= p

i
(
i
) for i < . If
we force in V
0
by
P

def
= limP

i
/(p

i

i
), Q

D
i
: i < ),
we obtain a universe V

in which T
i
: cf(i) =
+
) is an increasing sequence
of normal lters over , and T
def
=

iS

+
T
i
is a normal ultralter over .
For, in V
P
i
/(p

i

i
)
, we have that T
i
is an ultralter over , and cf() > ,
while the iteration is with (< )-supports and
<
= . Hence every subset
of in V

appears as an element of V
P
i
/(p

i

i
)
for some i, and so T is an
ultralter.
32
6
5
9


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







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


Also, for every i S

+
we have that A

i
T. Let T

be a P

-name for T
of V

. Let
E
def
=
_

_
< :
( < )( (, ))[

= and
T

T()
V
P

0
is a P

/(p

)-name
and p
+1
() = T T()
V
P

0
]
_

_
.
Hence E is a club of . Let E S

++
be larger than
+++
. Force with
P

, so obtaining V
1
in which 2

+
2

+++
, as each coordinate of
P

adds a subset of , and cardinals are preserved. In V

1
force with the
Prikry forcing for T

def
=

iS

+
T
i
. Let W
def
= V
1
[Pr(T

)]. For i S

+
, let
Gr

i
be a graph obtained in W satisfying the conditions of Conclusion 2.22
with T

in place of T

and T
i
in place of T. Let C be a club of of order
type
++
, and let g be its increasing enumeration.
We claim that W is as required, and that
Gr

g(i)
: i <
++
& cf(g(i)) =
+

are universal for graphs of size

+
. Clearly the conality of in W is
0
and is a strong limit. Suppose that Gr is a graph on
+
in W and let Gr

be a Pr(T

)-name for it. Hence, there is a i <

++
with cf(g(i)) =
+
such
that Gr

is a Pr(T
g(i)
)-name for a graph on
+
. The conclusion follows by
the choice of Gr

i
.
2.1
33
6
5
9


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







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


References
[ChKe] C. C. Chang and H. J. Keisler, Model theory, North-Holland, 1st
edition 1973, latest edition 1990.
[FuKo] Z. F uredi and P. Komjath, Nonexistence of universal graphs with-
out some trees, Combinatorica 17 (1997), pp. 163-171.
[GrSh 174] R. Grossberg and S. Shelah, On universal locally nite groups,
Israel Journal of Mathematics, 44 (1983), pp. 289-302.
[DjSh 614] M. Dzamonja and S. Shelah, On the existence of universals, sub-
mitted.
[DjSh 710] M. Dzamonja and S. Shelah, On properties of theories which pre-
clude the existence of universal models, submitted.
[GiSh 597] M. Gitik and S. Shelah, On densities of box products, Topology
and its Applications 88 (3) (1998), pp. 219-237.
[KaReSo] A. Kanamori, W. Reindhart and R. Solovay, Strong axioms of in-
nity and elementary embeddings, Annals of Mathematical Logic
13 (1978), pp. 73-116.
[Kj] M. Kojman, Representing embeddability as set inclusion, Journal
of LMS (2nd series), No.185, (58) (2) (1998), pp. 257-270.
[KoSh 492] P. Komjath and S.Shelah, Universal graphs without large cliques,
Journal of Combinatorial Theory (Series B) (1995), pp. 125-135.
[KjSh 409] M. Kojman and S. Shelah, Non-existence of Universal Orders in
Many Cardinals, Journal of Symbolic Logic, 57 (1992), pp. 875-
891.
[KjSh 447] M. Kojman and S. Shelah, The universality spectrum of stable un-
superstable theories, Annals of Pure and Applied Logic 58 (1992),
pp. 57-92.
34
6
5
9


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







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


[La] R. Laver, Making the supercompactness of indestructible under
-directed closed forcing, Israel Journal of Mathematics, 29 (4)
(1978), pp. 385-388.
[Ma 1] M. Magidor, On the singular cardinals problem I, Israel Journal
of Mathematics 28 (1977), pp. 1-31.
[Ma 2] M. Magidor, On the singular cardinals problem II, Annals of Math-
ematics 106 (1977), pp. 517-549.
[Ma 3] M. Magidor, Changing conality of cardinals, Fundamenta Math-
ematicae, XCIX (1978), pp. 61-71.
[MkSh 274] A. Mekler and S. Shelah, Uniformization Principles, The Jour-
nal of Symbolic Logic, 54 (2) (1989), pp. 441-459.
[Ra] L. Radin, Adding closed conal sequences to large cardinals, An-
nals of Mathematical Logic 22 (1982), pp. 243-261.
[Rd] R. Rado, Universal graphs and universal functions, Acta Arith-
metica (9) (1964), pp. 331-340.
[Sh -f] S. Shelah, Proper and improper forcing, 2nd edition, Springer,
Perspectives in Mathematical Logic, 1998.
[Sh 80] S. Shelah, A weak generalization of MA to higher cardinals, Israel
Journal of Mathematics, 30 (1978) pp. 297-308.
[Sh 93] S. Shelah, Simple Unstable Theories, Annals of Mathematical
Logic, 19 (1980) pp. 177-204.
[Sh 175a] S. Shelah, Universal graphs without instances of CH: revisited Is-
rael Journal of Mathematics, 70 (1990), pp. 69-81.
[Sh 457] S. Shelah, the Universality Spectrum: Consistency for more
Classes, in Combinatorics, Paul Erd os is Eighty, vol. 1,
pp. 403-420, Bolyai Society Mathematical Studies, 1993. Pro-
ceedings of the Meeting in honour of Paul Erd os, Ket-
zhely, Hungary 7. 1993; An elaborated version available from
http://www.math.rutgers.edu/shelarch
35
6
5
9


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







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


[Sh 500] S. Shelah, Toward classifying unstable theories, Annals of Pure
and Applied Logic 80 (1996), pp. 229-255.
[Sh 546] S. Shelah, Was Sierpinski right IV, Journal of Symbolic Logic, 65
(2000), no.3, 1031-1054.
36