2

8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


COLLOQUIA MATHEMATICA SOCIETATIS J

ANOS BOLYAI
60. SETS, GRAPHS AND NUMBERS, BUDAPEST (HUNGARY), 1991
Strong Partition Realations Below the
Power Set: Consistency
Was Sierpinski Right? II.
S. SHELAH

We continue here [Sh276] (see the introduction there) but we do not

relay on it. The motivation was a conjecture of Galvin stating that 2

2
+
2
[
1
]
n
h(n)
is consistent for a suitable h : . In section 5
we disprove this and give similar negative results. In section 3 we prove
the consistency of the conjecture replacing
2
by 2

, which is quite large,

starting with an Erd os cardinal. In section 1 we present iteration lemmas
which needs when we replace by a larger and in section 4 we generalize
a theorem of Halpern and Lauchli replacing by a larger .
0. Preliminaries
Let <

be a well ordering of H(), where H() = x : the transitive closure

of x has cardinality < , agreeing with the usual well-ordering of the

Publication no 288, summer 86. The author would like to thank the United States
Israel Binational Science Foundation for partially supporting this research.
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


2
ordinals. P (and Q, R) will denote forcing notions, i.e. partial orders with
a minimal element =
P
.
A forcing notion P is -closed if every increasing sequence of members
of P, of length less than , has an upper bound.
If P H(), then for a sequence p = p
i
: i < ) of members of P let

p
def
= supj

:
j
: j < j

has an upper bound in P and dene the

canonical upper bound of p, &p as follows:
(a) the least upper bound of p
i
: i <

in P if there exists such an

element,
(b) the <

-rst upper bound of p if (a) cant be applied but there is such,

(c) p
0
if (a) and (b) fail, > 0,
(d)
P
if = 0.
Let p
0
&p
1
be the canonical upper bound of p

: < 2).
Take [a]

= b a : [b[ = and [a]

<
=

<
[a]

.
For sets of ordinals, A and B, dene H
OP
A,B
as the maximal order
preserving bijection between initial segments of A and B, i.e, it is the
function with domain A : otp( A) < otp(B), and H
OP
A,B
() = if
and only if A, B and otp( A) = otp( B).
Denition 0.1
+
()
<

holds provided whenever F is a function from

[]
<
to , C is a club then there is A C of order type such that
[w
1
, w
2
[A]
<
, [w
1
[ = [w
2
[ F(w
1
) = F(w
2
)].
Denition 0.2 []
n
,
if for every function F from []
n
to there is
A of order type such that F(w) : w [A]
n
has power .
Denition 0.3 A forcing notion P satises the Knaster condition (has
property K) if for any p
i
: i <
1
P there is an uncountable A
1
such that the conditions p
i
and p
j
are compatible whenever i, j A.
1. Introduction
Concerning 1.11.3 see Shelah [Sh80], Shelah and Stanley [ShSt154, 154a].
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


3
Denition 1.1. A forcing notion Q satises

where is a limit ordinal

< , if player I has a winning strategy in the following game:
Playing: the play nishes after moves.
in the
th
the move:
Player I if ,= 0 he chooses q

: <
+
) such that q

Q and
( < )( <
+
)p

and he chooses a regressive

function f

:
+

+
(i.e. f

(i) < 1 +i); if = 0 let

q

=
Q
, f

= .
Player II he chooses p

: <
+
) such that q

Q.
The outcome: Player I wins provided whenever < < <
+
, cf() =
cf() = and
<
f

() = f

() the set p

: < p

: < has
an upper bound in Q.
Denition 1.2. We call P
i
, Q
j
: i i(), j < i()) a

-iteration
provided that:
(a) it is a (< )-support iteration ( is a regular cardinal)
(b) if i
1
< i
2
i(), cf i
1
,= then P
i
2
/P
i
1
satises

.
The Iteration Lemma 1.3. If

Q = P
i
, Q
j
: i i(), j < i()) is a
(< )-support iteration, (a) or (b) or (c) below hold, then it is a

-iteration.
(a) i() is limit and

Qj() is a

-iteration for every j() < i().

(b) i() = j() +1,

Qj() is a

-iteration and Q
j()
satises

in V
P
j()
.
(c) i() = j() + 1, cf j() =
+
,

Q j() is a

-iteration and for every

successor i < j(), P
i()
/P
i
satises

.
Proof. Left to the reader (after reading [Sh80] or [ShSt154a]).
Theorem 1.4. Suppose =
<
< < , and is a strongly inaccessible
k
2
2
-Mahlo cardinal, where k
2
2
is a suitable natural number (see 3.6(2) of
[Sh289]), and assume V = L for the simplicity. Then for some forcing
notion P:
(a) P is -complete, satises the
+
-c.c., has cardinality , and V
P
[=
2

= .
(b)
P
[
+
]
2
3
and even [
+
]
2
,2
for < .
(c) if =
0
then MA

.
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


4
(d) if >
0
then:
P
for every forcing notion Q of cardinality , -
complete satisfying

, and for any dense sets D

i
Q for i < i
0
< , there
is a directed G Q,
i
G D
i
,= .
As the proof is very similar to [Sh276], (particularly after reading section
3) we do not give details. We shall dene below just the systems needed to
complete the proof. More general ones are implicit in [Sh289].
Convention 1.5. We x a one to one function Cd = Cd
,
from
>
onto
.
Remark. Below we could have otp(B
x
) =
+
+ 1 with little change.
Denition 1.6. Let < < , =
<
, =
<
, =
<
.
1) We call x a (, , , )-precandidate if x = a
x
u
: u I
x
) where for some
set B
x
(unique, in fact):
(i) I
x
= s : s B
x
, [x[ 2,
(ii) B
x
is a subset of of order type
+
,
(iii) a
x
u
is a subset of of cardinality closed under Cd,
(iv) a
x
u
B
x
= u,
(v) a
x
u
a
x
v
a
x
uv
,
(vi) if u, v I
x
, [u[ = [v[ then a
x
u
and a
x
v
have the same order type (and
so H
OP
a
x
u
,a
x
v
maps a
x
u
onto a
x
v
),
(vii) if u

, v

I
x
for = 1, 2, [u
1
[ = [v
1
[, [u
2
[ = [v
2
[, [u
1
u
2
[ = [v
1
v
2
[,
H
OP
a
x
u
1
a
x
u
2
,a
x
v
1
a
x
v
2
maps u

onto v

for = 1, 2 then H
OP
a
x
u
1
,a
x
v
1
and
H
OP
a
x
u
2
,a
x
v
2
are compatible.
2) We say x is a (, , , )-candidate if it has the form M
x
u
: u I
x
)
where
() (i) [M
x
u
[ : u I
x
) is a (, , , )-precandidate (with B
x
def
= I
x
)
(ii) L
x
is a vocabulary with -many < -ary placespredicates and
function symbols,
(iii) each M
x
u
is an L
x
-model,
(iv) for u, v I
x
, [u[ = [v[, M
x
u
([M
x
u
[ [M
x
v
[) is a model, and in fact
an elementary submodel of M
x
v
, M
x
u
and M
x
uv
.
() () for u, v I
x
, [u[ = [v[, the function H
OP
|M
x
u
|,|M
x
v
|
is an isomorphism
from M
x
u
onto M
x
v
.
3) The set A is a (, , , )-system if
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


5
(A) each x A is a (, , , )-candidate,
(B) guessing: if L is as in (2)()(ii), M

is an L-model with universe

then for some x A, s B
x
M
x
s
M

.
Denition 1.7. 1) We call the system A disjoint when:
() if x ,= y are from A and otp([M
x

[) otp([M
y

[) then for some B

1
B
x
,
B
2
B
y
we have
a) [B
1
[ +[B
2
[ <
+
b) the sets
_
[M
x
s
[ : s [B
x
B
1
]
2

and
_
[M
y
s
[ : s [B
y
B
2
]
2

have intersection M
y

.
2) We call the system A almost disjoint when:
() if x, y A, otp([M
x

[) otp([M
y

[) then for some B

1
B
x
,
B
2
B
y
we have:
(a) [B
1
[ +[B
2
[ <
+
,
(b) if s [B
x
B
1
]
2
, t [B
y
B
2
]
2
then [M
x
s
[ [M
x
t
[ [M
y

[.
2. Introducing the partition on trees
Denition 2.1. Let
1) Per(
>
2) = T : where
(a) T
>
2, ) T,
(b) ( T) ( < lg()) T,
(c) if T

2, < < then for some

T

2, ,
(d) if T then for some , ,
0) T, 1) T,
(e) if

2, < is a limit ordinal and
: < T then T.
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


6
2) Per
f
(
>
2) =
_
T Per(
>
2) : if < and
1
,
2


2 T, then
[
1

=0

1
) T
1

=0

2
) T]
_
.
3) Per
u
(
>
2) = T Per(
>
2) : if < ,
1
,=
2
from

2 T,
then
1

=0
2

m=1

m
) / T.
4) For T Per(
>
2) let limT =

2 : ( < ) T.
5) For T Per
f
(
>
2) let clp
T
: T
>
2 be the unique one-to-one
function from sp(T)
def
= T : 0) T, 1) T onto
>
2,
which preserves and lexicographic order.
6) Let SP(T) = lg() : sp(T), sp(, ) = mini : (i) ,= (i) or i =
lg() or i = lg().
Denition 2.2. 1) For cardinals , and n < and T Per(
>
2) let
Col
n

(T) = d : d is a function from

<
[

2]
n
T to . We will write
d(
0
, . . . ,
n1
) for d(
0
, . . . ,
n1
).
2) Let <

denote a well ordering of

2 (in this section it is arbitrary). We
call d Col
n

(T) end-homogeneous for <

: < ) provided that: if

< are from SP(T),
0
, . . . ,
n1


2 T,

: < n) are
pairwise distinct and
_
,m
[

<

<

m
] then
d(
0
, . . . ,
n1
) = d(
0
, . . . ,
n1
).
3) Let Eh Col
n

(T) = d Col
n

(T) : d is end-homogeneous (for some

<

: < )).
4) For
0
, . . . ,
n1
,
0
, . . . ,
n1
from
>
2, we say =
0
, . . . ,
n1
) and
=
o
, . . . ,
n1
) are strongly similar for <

: < ) if:
(i) lg(

) = lg(

)
(ii) sp(

,
m
) = sp(

,
m
)
(iii) if
1
,
2
,
3
,
4
< n and = sp(

1
,

2
) then

3
<

3
<

4
and

3
() =

3
()
5) For
a
0
, . . . ,
a
n1
,
b
0
, . . . ,
b
n1
from
>
2 we say
a
=
a
0
, . . . ,
a
n1
) and

b
=
b
0
, . . . ,
b
n1
) are similar if the truth values of (i)(iii) below doe
not depend on t a, b for any (1), (2), (3), (4) < n:
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


7
(i) lg(
t
(1)
) < lg(
t
(2)
)
(ii) sp(
t
(1)
,
t
(2)
) < sp(
t
(3)
,
t
(4)
)
(iii) for = sp(
t
(1)
,
t
(2)
),

t
(3)
<

t
(4)

and

t
(3)
() = 0.
6) We say d Col
n

(T) is almost homogeneous [homogeneous] on T

1
T
(for <

: < )) if for every SP(T

1
), , [

2]
n
T
1
which are
strongly similar [similar] we have d( ) = d( ).
7) We say <

: < ) is nice to T Per(

>
2), provided that: if
< are from SP(T), (, ) SP(T) = ,
1
,=
2


2 T,
[
1
<

2
or
1
=
2
,
1
() <
2
()] then
1
<

2
.
Denition 2.3. 1) Pr
eht
(, n, ) means: for every d Col
n

(
>
2) for some
T Per(
>
2), d is end homogeneous on T.
2) Pr
aht
(, n, ) means for every d Col
n

(
>
2) for some T Per(
>
2), d
is almost homogeneous on T.
3) Pr
ht
(, n, ) means for every d Col
n

(
>
2) for some T Per(
>
2), d
is homogeneous on T.
4) For x eht, aht, ht, Pr
f
x
(, n, ) is dened like Pr
x
(, n, ) but we
demand T Per
f
(
>
2).
5) If above we replace eht, aht, ht by ehtn, ahtn, htn, respectively, this
means <

: < ) is xed apriori.

6) Replacing n by < , by =

: < ) for
0
, means that
d
n
: n < ) are given, d
n
Col
n

(
>
2) and the conclusion holds for
all d
n
(n < ) simultaneously. Replacing by < means that the
assertion holds for every
1
< .
Denition 2.4. 1) Pr
aht
(, n, (1), (2)) means: for every d Col
n
(1)
(
>
2) for some T Per(
>
2) and <

: < ) for every

2]
n
T :
SP(T),
_
d( ) :
_
[

2]
n
T
1
: SP((T
1
),
and are strongly similar for <

: < )
_
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


8
has cardinality < (2).
2) Pr
ht
(, n, (1), (2)) is dened similarly with similar instead of
strongly similar.
3) Pr
x
_
, < ,
1

: < )
2

: < )
_
, Pr
f
x
(, n, (1), (2)), Pr
f
x
(, <

0
,
1
,
2
) are dened in the same way.
There are many obvious implications.
Fact 2.5. 1) For every T Per(

>2) there is a T
1
T, T
1
Per
u
(
>
2).
2) In dening Pr
f
x
(, n, ) we can demand T T
0
for any T
0
Per
f
(
>
2),
similarly for Pr
f
x
(, < , ).
3) The obvious monotonicity holds.
Claim 2.6. 1) Suppose is regular,
0
and Pr
f
eht
(, n, < ). Then
Pr
f
aht
(, n, < ) holds.
2) If is weakly compact and Pr
f
aht
(, n, < ), < , then Pr
f
ht
(, n, < )
holds.
3) If is Ramsey and Pr
f
aht
(, <
0
, < ), < , then Pr
f
ht
(, <
0
, < ).
4) If = , in the nice version, the orders <

: < ) disappear.
Proof. : Check it.
The following theorem is a quite strong positive result for = .
Halpern Lauchli proved 2.7(1), Laver proved 2.7(2) (and hence (3)), Pincus
pointed out that Halpern Lauchlis proof can be modied to get 2.7(2), and
then Pr
f
eht
(, n, < ) and (by it) Pr
f
ht
(, n, < ) are easy.
Theorem 2.7. 1) If d Col
n

(
>
2), <
0
, then there are T
0
, . . . , T
n1

Per
f
(
>
2) and k
0
< k
1
< . . . < k

< . . . and s < such that for every

< : if
0
T
0
,
1
T
1
, . . . ,
n1
T
n1
,
_
m<n
lg(
m
) = k

, then
d(
0
, . . . ,
n1
) = s.
2) We can demand in (1) that
SP(T

) = k
0
, k
1
, . . .
3) Pr
f
htn
(, n, ) for <
0
.
4) Pr
f
htn
_
, <
0
,
1
n
: n < ),
2
n
: n < )
_
if
1
n
<
0
and
2
n
: n < )
diverge to innity.
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


9
Denition 2.8. Let d be a function with domain [A]
n
, A be a set of
ordinals, F be a one-to-one function from A to
()
2, <

be a well ordering
of

2 for () such that F() <

F() < , and be a cardinal.

1) We say d is (F, )-canonical on A if for any
1
< <
n
A,

_
d(
1
, . . . ,
n
) : F(
1
), . . . , F(
n
)) similar to
F(
1
), . . . , F(
n
))
_

.
2) We dene almost (F, )-canonical similarly using strongly similar
instead of similar.
3. Consistency of a strong partition below the continuum
This section is dedicated to the proof of
Theorem 3.1. Suppose is the rst Erd os cardinal, i.e. the rst such that
(
1
)
<
2
. Then, if A is a Cohen subset of , in V [A] for some
1
c.c.
forcing notion P of cardinality ,
P
MA

1
(Knaster) + 2

0
= and:
1.)
P
[
1
]
n
h(n)
for suitable h : (explicitly dened below).
2.) In V
P
for any colorings d
n
of , where d
n
is n-place, and for any diver-
gent
n
: n < ) (see below), there is a W , [W[ =
1
and a function
F : W

2 such that: d
n
is (F,
n
) canonical on W for each n.
(See denition 2.8 above.)
Remark 3.2. h(n) is n! times the number of u [

2]
n
satisfying (if

1
,
2
,
3
,
4
u are distinct then sp(
1
,
2
), sp(
3
,
4
) are distinct) up to
strong similarity for any nice <

: < ).
2) A sequence
n
: n < ) is divergent if m k n k
n
m.
Notation 3.3. For a sequence a =
i
, e

i
: i < ), we call b closed if
(i) i b a
i
b
(ii) if i < , e

i
= 1 and sup(b i) = i then i b.
Denition 3.4. Let K be the family of

Q = P
i
, Q

j
, a
j
, e

j
: j < , i )
such that
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


10
(a) a
i
i, [a
i
[
1
,
(b) a
i
is closed for a
j
, e

j
: j < i), e

i
0, 1, and [e

i
= 1 cf i =
1
]
(c) P
i
is a forcing notion, Q

j
is a P
j
-name of a forcing notion of power
1
with minimal element or
j
and for simplicity the underlying set of
Q

j
is [
1
]
<
0
(we do not lose by this).
(d) P

= p : p is a function whose domain is a nite subset of and for

i dom(p),
P
i
f(i) Q

i
with the order p q if and only if for
i dom(p), qi
P
i
p(i) q(i).
(e) for j < i, Q

j
is a P
j
-name involving only antichains contained in
p P
j
: dom(p) a
j
.
For p P
i
, j < i, j , domp we let p(j) = . Note for p P
i
, j i,
pj P
j
Denition 3.5. For

Q K as above (so = lg(

Q)):
1) for any b closed for a
i
, e

i
: i < ) we dene P
cn
b
[by
simultaneous induction on ]:
P
cn
b
= p P

: domp b, and for i domp, p(i) is a canonical name

i.e., for any x, p P
cn
a
i
: p
P
i
p(i) = x or p
P
i
p(i) ,= x is a predense
subset of P
i
.
2) For

Q as above, = lg(

Q), take

Q = P
i
, Q

j
, a
j
: i , j < ) for
and the order is the order in P

(if ,

Q =

Q).
3) b closed for

Q means b closed for a
i
, e

i
: i < lg

Q).
Fact 3.6. 1) if

Q K then

Q K.
2) Suppose b c lg(

), b and c are closed for

Q K.
(i) If p P
cn
c
then pb P
cn
b
.
(ii) If p, q P
cn
c
and p q then pb qc.
(iii) P
cn
c
P

. 3) lg

Q is closed for

Q.
4) if

Q K, = lg

Q then P
cn

is a dense subset of P

.
5) If b is closed for

Q, p, q P
cn
lg

Q
, p q in P
lg

Q
and i domp then qa
i

P
i
p(i) q(i) hence
P
cn
a
i
p(i)
Q
i
q(i).
Denition 3.7. Suppose W = (W, ) is a nite partial order and

Q K.
1) IN
W
(

Q) is the set of

b-s satisfying ()() below:
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


11
()

b = b
w
: w W) is an indexed set of

Q-closed subsets of lg(

Q),
() W [= w
1
w
2
b
w
1
b
w
2
,
() b
w
1
b
w
2
, w
1
w, w
2
w then (u W) b
u
u w
1
u w
2
.
We assume

b codes (W, ).
2) For

b IN
W
(

Q), let

Q[

b]
def
= p
w
: w W) : p
w
P
cn
b
w
, [W [= w
1
w
2
p
w
2
b
w
1
= p
w
1
]
with ordering

Q[

b] [= p
1
p
2
i
_
wW
p
1
w
p
2
w
.
3) Let K
1
be the family of

Q K such that for every lg(

Q) and (

Q)-
closed b, P

and P

/P
cn
b
satisfy the Knaster condition.
Fact 3.8. Suppose

Q K
1
, (W, ) is a nite partial order,

b IN
W
(

Q)
and p

Q[

b].
1) If w W, p
w
q P
cn
b
w
then there is r

Q[

b], q r
w
, p r, in fact
r
u
() =
_

_
p
u
() if Domp
u
Domq
p
u
() & q() if b
u
Domq and for some v W,
v u, v w and b
v
p
u
() if b
u
domq but the previous case fails
2) Suppose (W
1
, ) is a submodel of (W
2
, ), both nite partial orders,

b
l
IN
W
l
(

Q),

b
1
w
=

b
2
w
for w W
1
.
() If q

Q[

b
2
] then q
w
: w W
1
)

Q[

b
1
].
() If p

Q[

b
1
] then there is q

Q[

b
2
], qW
1
= p, in fact q
w
() is p
u
() if
u W
1
, b
u
, u w, provided that
() if w
1
, w
2
W
1
, w W
2
, w
1
w, w
2
w and b
w
1
b
w
2
then for
some v W
1
, b
v
, v w
1
, v w
2
.
(this guarantees that if there are several us as above we shall get the same
value).
3) If

Q K
1
then

Q[

b] satises the Knaster condition. If is the minimal

element of W (i.e. u W W [= u) then

Q[

b]/P
cn
b

also satises the

Knaster condition and so

Q[

b], when we identify p P

cn
b
with p : w
W).
Proof. 1) It is easy to check that each r
u
() is in P
cn
b
u
. So, in order to prove
r

Q[

b], we assume W [= u
1
u
2
and has to prove that r
u
2
b
u
1
= r
u
1
. Let
b
u
1
.
First case: , Dom(p
u
1
) Domq.
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


12
So , Dom(r
u
1
) (by the denition of r
u
1
) and , Domp
u
2
(as
p

Q[

b]) hence , (Domp

u
2
) (Domq) hence , Dom(r
u
2
) by the
choice of r
u
2
, so we have nished.
Second case: Domp
u
1
Domq.
As p

Q[

b] we have p
u
1
() = p
u
2
(), and by their denition, r
u
1
() =
p
u
1
(), r
u
2
() = p
u
2
().
Third case: Domq and (v W) ( b
v
v u
1
v w). By the
denition of r
u
1
(), we have r
u
1
() = p
u
1
()&q(), also the same v witnesses
r
u
2
() = p
u
2
()&q(), (as b
v
v u
1
v w b
v
v u
2
v w)
and of course p
u
1
() = p
u
2
() (as p

Q[

b]).
Fourth case: Domq and (v W) ( b
v
v u
1
v w).
By the denition of r
u
1
() we have r
u
1
() = p
u
1
(). It is enough to prove
that r
u
2
() = p
u
2
() as we know that p
u
1
() = p
u
2
() (because p

Q[

b],
u
1
u
2
). If not, then for some v
0
W, b
v
0
v
0
u
2
v
0
w. But

b IN
W
(

Q), hence (see Def. 3.7(1) condition () applied with , w

1
, w
2
, w
there standing for , v
0
, u
1
, u
2
here) we know that for some v W,
v v v
0
v u
1
. As (W, ) is a partial order, v v
0
and v
0
w, we
can conclude v w. So v contradicts our being in the fourth case. So we
have nished the fourth case.
Hence we have nished proving r

Q[

b]. We also have to prove q r

w
,
but for Domq we have b
w
(as q P
cn
w
is on assumption) and
r
w
() = q() because r
w
() is dened by the second case of the denition
as (v W) ( b
w
v w v w), i.e. v = w.
Lastly we have to prove that p r (in

Q[

b]). So let u W, Domp

u
and we have to prove r
u

P

p
u
()
P

r
u
(). As r
u
() is p
u
() or
p
u
()&q() this is obvoius.
2) Immediate.
3) We prove this by induction on [W[.
For [W[ = 0 this is totally trivial.
For [W[ = 1, 2 this is assumed.
For [W[ > 2 x p
i

Q[

b] for i <
1
. Choose a maximal element v W and
let c =

b
w
: W [= w < v. Clearly c is closed for

Q.
We know that P
cn
c
, P
cn
b
v
/P
cn
c
are Knaster by the induction hypothesis.
We also know that p
i
v
c P
cn
c
for i <
1
, hence for some r P
cn
c
,
r A

def
=
_
i <
1
: p
i
v
c G

P
cn
c
_
is uncountable
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


13
hence
there is an uncountable A
1
A

such that
_
i, j A
1
p
i
v
, p
j
v
are compatible in P
cn
b
v
/ G

P
cn
c
_
.
Fix a P
cn
c
-name A

1
for such an A
1
.
Let A
2
=
_
i <
1
: q P
cn
c
, q i A

1
_
. Necessarily
[A
2
[ =
1
, and for i A
2
there is q
i
P
cn
c
, q
i
i A
1
, and
w.l.o.g. p
i
v
c q
i
. Note that p
i
v
&q
i
P
cn
c
.
For i A
2
let, r
i
be dened using 3.8(1) (with p
i
, p
i
v
&q
i
). Let W
1
=
W v,

b

= b
w
: w W
1
).
By the induction hypothesis applied to W
1
,

b

, r
i
W
1
, for i A
2
there is an uncountable A
3
A
2
and for i < j in A
3
, there is r
i,j

Q[

],
r
i
W
1
r
i,j
, and r
j
W
1
r
i,j
. Now dene r
i,j
c
P
cn
c
as follows: its domain
is
_
domr
i,j
w
: W [= w < v
_
, r
i,j
c
(domr
i,j
w
) = r
i,j
w
whenever W [= w < v.
Why is this a denition? As if W [= w
1
v w
2
v, b
w
1
b
w
2
then for some u W, u w
1
u w
2
and u. It is easy to check that
r
i,j
c
P
cn
c
. Now r
i,j
c

P
cn
c
p
i
b
v
, p
j
b
v
are compatible in P
cn
b
v
/P
cn
c
.
So there is r P
cn
b
v
such that r
i,j
c
r, p
i
b
v
r, p
j
b
v
r. As in part (1) of
3.8 we can combine r and r
i,j
to a common upper bound of p
i
, p
j
in

Q[

b].
Claim 3.9. If e = 0, 1 and is a limit ordinal, and P
i
, Q

i
,
i
, e

i
(i < ) are
such that for each < ,

Q

= P
i
, Q

j
,
j
, e

j
: i , j < ) belongs to
K

, then for a unique P

,

Q = P
i
, Q

j
,
j
, e

j
: i , j < ) belongs to K

.
Proof. We dene P

by (d) of Denition 3.4. The least easy problem is to

verify the Knaster conditions (for

Q K
1
). The proof is like the preservation
of the c.c.c. under iteration for limit stages.
Convention 3.9A. By 3.9 we shall not distinguish strictly between P
i
, Q

j
,

j
, e

j
: i , j < ) and P
i
, Q

i
,
i
, e

i
: i < ).
Claim 3.10. If

Q K

, = lg(

Q), a is closed for

Q, [a[
1
, Q

1
is
a P
cn
a
-name of a forcing notion satisfying (in V
P

) the Knaster condition,

its underlying set is a subset of [
1
]
<
0
then there is a unique

Q
1
K

,
lg(

Q
1
) = + 1, Q
1

= Q

,

Q =

Q.
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


14
Proof. Left to the reader.
Proof of Theorem 3.1.
A Stage: We force by K
1
<
=
_

Q K
1
: lg(

Q) < ,

Q H()
_
ordered by
being an initial segment (which is equivalent to forcing a Cohen subset of
). The generic object is essentially

Q

K
1

, lg(

) = , and then we force

by P

= lim

Q

. Clearly K

<
is a -complete forcing notion of cardinality
, and P

satises the c.c.c. Clearly it suces to prove part (2) of 3.1.

Suppose d

n
is a name of a function from []
n
to k

n
for n < ,

n
< ,

n
: n < ) diverges (i.e. m k n k
n
m) and for some

Q
0
K
1
<
.

Q
0

K
1
<
there is p P

[p
P

n
: n < ) is a
counterexample to (2) of 3.1].
In V we can dene

: < ),

Q

K
1
<
, <

Q

=

Q

lg(

),
in

Q
+1
, e

lg(

)
= 1,

Q
+1
forces (in K
1
<
) a value to p and the P

-names
d

n
,

n
, k

n
for n < , i.e. the values here are still P

-names. Let

Q

be the limit of the

Q

-s. So

Q

K
1
, lg(

) = ,

Q

= P

i
, Q

j
,

j
, e

j
:
i , j < ), and the P

-names d

n
,

n
, k

n
are dened such that in V
P

,
d

n
,

n
, k

n
contradict (2) (as any P

-name of a bounded subset of is a

P

lg(

)
-name for some < ).
B Stage: Let =
+
and <

be a well-ordering of H(). Now we can apply

(
1
)
<
2
to get , B, N
s
(for s [B]
<
0
) and h
s,t
(for s, t [B]
<
0
, [s[ =
[t[) such that:
(a) B , otp(B) =
1
, supB = ,
(b) N
s
(H(), , <

),

Q

N
s
, d

n
, k

n
: n < ) N
s
,
(c) N
s
N
t
= N
st
,
(d) N
s
B = s,
(e) if s = t , t [B]
<
0
then N
s
is an initial segment of N
t
,
(f) h
s,t
is an isomorphism from N
t
onto N
s
(when dened)
(g) h
t,s
= h
1
s,t
(h) p
0
N
s
, p
0

P

n
,

n
, k

n
: n <) is a counterexample,
(i)
1
N
s
, [N
s
[ =
1
and if N
s
, cf >
1
then cf(sup( N
s
)) =

1
.
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


15
Let

Q =

Q

, P = P

and P
a
= P
cn
a
(for

Q), where a is closed for

Q.
Note: P

N
s
= P

N
s
= P
sup N
s
N
s
= P
s
N
s
. Note also N
s
a

N
s
.
C Stage: It suces to show that we can dene Q

in V
P

which forces
a subset W of B of cardinality
1
and F

: W

2 which exemplify the
desired conclusion in (2), and prove that Q

satises the
1
-c.c.c. (in V
P

(and has cardinality

1
)) and moreover (see Denitions 3.4 and 3.7(3)) we
also dene a

=

s[B]
<
0
N
s
, e

= 1,

Q

=

QP

, Q

, a

, e

) and prove

K
1
.
We let d

(u) = d

|u|
(u).
Let F :
1


2 be one-to-one such that
>
2

1
<
1
[F()].
(This will not be the needed F

, just notation).
For s, t [B]
<
0
, we say s
n
F
t if [s[ = [t[ and s, t[ =
h
s,t
() F()n = F()n]. Let I
n
= I
n
(F) = s [B]
<
0
: ( ,= s),
[F()n ,= F()n].
We dene R
n
as follows: a sequence p
s
: s I
n
) R
n
if and only if
(i) for s I
n
, p
s
P

N
s
,
(ii) for some c
s
we have p
s
d

(s) = c
s
,
(iii) for s, t I
n
, s
n
F
t h
s,t
(p
t
) = p
s
,
(iv) for s, t I
n
, p
s
N
st
= p
t
N
st
.
R

n
is dened similarly omitting (ii).
For x = p
s
: s I
n
) let n(x) = n, p
x
s
= p
s
, and (if dened)
c
x
s
= c
s
. Note that we could replace x R
n
by a nite subsequence.
Let R =

n<
R
n
, R =

n<
R

n
. We dene an order on R : x y if
and only if n(x) n(y), and [s I
n(x)
t I
n(y)
s t p
x
s
p
y
t
].
D Stage: Note the following facts::
D() Subfact: If x R

n
, t I
n
and p
x
t
p
1
P

N
t
, then there is y
such that x y R

n
, p
y
t
= p
1
.
Proof. We let for s I
n
p
y
s
def
= &
_
h
s
1
,t
1
(p
1
N
t
1
) : s
1
s, t
1
t, s
1

n
F
t
1
_
&p
x
s
.
(This notation means that p
y
s
is a function whose domain is the union
of the domains of the conditions mentioned, and for each coordinate we
take the canonical upper bound, see preliminaries.) Why is p
y
s
well dened?
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


16
Suppose N
s
(for N
s
, clearly p
y
s
() =

), s

s, t

t,
s

n
F
t

for = 1, 2 and Dom

_
h
s

,t

(p
1
N
t

)
_
, and it suces to show
that p
x
s
(), h
s
1
,t
1
(p
1
N
t
1
)(), h
s
2
,t
2
(p
1
N
t
2
)() are pairwise comparable.
Let u =

v [B]
<
0
: N
v
, necessarily u s
1
s
2
, and let
u

= h
1
s

,t

(u). As s

, t

, t I
n
, s

n
F
t

and u

t, necessarily
u
1
= u
2
. Thus
def
= h
1
u,v
() = h
1
s

,t

() and so the last two conditions are

equal.
Now p
x
s
() = p
x
u
() = h
u,v
(p
x
s
()) h
s

,t

((p
x
t
N
t

)()) =
_
h
s

,t

(p
x
t

N
t

)
_
().
We leave to the reader checking the other requirements.
D() Subfact: If x R

n
, t I then

p
x
s
: s I
n
, s t (as union of
functions) exists and belongs to P

N
t
.
Proof. See (iv) in the denition of R

n
.
D() Subfact: If x y, x R
n
, y R

n
, then y R
n
.
Proof. Check it.
D() Subfact: If x R

n
, n < m, then there is y R
m
, x y.
Proof. By subfact D() we can nd x
1
= p
1
t
: t I
m
) inR

m
with
x x
1
. Using repeatedly subfact D() we can increase x
1
(nitely many
times) to get y R
m
.
D() Subfact: If x R

n
, s, t I
n
, s
n
F
t, p
x
s
r
1
P

N
s
, p
x
t
r
2

P

N
t
, ( t) [F()(n) ,= (F(h
s,t
()))(n)] ( or just p
x
s
1
s
1
= h
s,t
(p
x
t
1
t
1
)
where t
1
def
= t : F()(n) = (F(h
s,t
()))(n), s
1
def
= h
s,t
() : t
1
),
then there is y R
n+1
, x y such that r
1
= p
y
s
and r
2
= p
y
t
.
Proof. Left to the reader.
E Stage

:

We will have T
>
2 gotten by 2.7(2) and then want to get a subtree with as
few as possible colors, we can nd one isomorphic to
>
2, and there restrict ourselves to
nT

n
.
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


17
We dene: T

k

2
k

2 by induction on k as follows:
T

0
=), 1)
T

k+1
= : T

k
or 2
k
< lg() 2
k+1
, 2
k
T

k
and
[2
k
i < 2
k+1
(i) = 1] i = 2
k
+ (

m<2
k
(i)2
m
)].
We dene
Tr Emb(k, n) =
_
h : h a is function from T

k
into
n
2 such that
for , T

k
:
[ = h() = h()]
[ h() h()]
[lg() = lg() lg(h() = lg(h()]
[ = i) (h())[lg(h())] = i]
[ lg() =
k
2 lg(h()) = n]
_
.
T(k, n) =Rang h : h Tr Emb(k, n),
T(, n) =
_
k
T(k, n),
T(k, ) =
_
k
T(k, n).
For T T(k, ) let n(T) be the unique n such that T T(k, n) and let
B
T
= B : F()n(T) is a maximal member of T,
fs
T
=
_
t B
T
: t t ,= n(T) ,= n(T),

T
=
_
p
s
: s fs
T
) : p
s
P N
s
, [s t s, t fs
T
p
s
= p
t
N
s
]
_
.
Let further

k
=
_

T
: T T(k, )
=
_
k

k
.
For p , n
p
= n(p), T
p
are dened naturally.
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


18
For p, q , p q i n
p
n
q
and for every s fs
T
p
we have p
s
q
s
.
F Stage: Let g

: , g

N
s
, g

grows fast enough relative

n
: n < ).
We dene a game Gm. A play of the game lasts after moves, in the n
th
move player I chooses p
n

n
and a function h
n
satisfying the restrictions
below and then player II chooses q
n

n
, such that p
n
q
n
(so T
p
n
= T
q
n
).
Player I loses the play if sometimes he has no legal move; if he never loses,
he wins. The restrictions player I has to satisfy are:
(a) for m < n, q
m
p
n
, p
n
s
forces a value to g

(n + 1),
(b) h
n
is a function from [B
T
p
n
]
g(n)
to ,
(c) if m < n h
n
, h
m
are compatible,
(d) If m < n, < g(m), s [B
T
p
n
]

, then p
n
s
d

(s) = h
n
(s),
(e) Let s
1
, s
2
Domh
n
. Then h
n
(s
1
) = h
n
(s
2
) whenever s
1
, s
2
are
similar over n which means:
(i)
_
F
_
H
OP
s
2
,s
1
()
_
_
n[p
n
] =
_
F()
_
n[p
n
] for s
1
,
(ii) H
OP
s
2
,s
1
preserves the relations sp
_
F(
1
), F(
2
)
_
< sp
_
F(
3
),
F(
4
)
_
and F(
3
)
_
sp(F(
1
), F(
2
))
_
= i (in the interesting
case
3
,=
1
,
2
implies i = 0).
G Stage/Claim: Player I has a winning strategy in this game.
Proof. As the game is closed, it is determined, so we assume player II has
a winning strategy , and eventually we shall get a contradiction. We dene
by induction on n, r
n
and
n
such that
(a) r
n
R
n
, r
n
r
n+1
,
(b)
n
is a nite set of initial segments of plays of the game,
(c) in each member of
n
player II uses his winning strategy,
(d) if y belongs to
n
then it has the form p
y,
, h
y,
, q
y,
: m(y)); let
h
y
= h
y,n
y
and T
y
= T
q
y
,m(y)
; also T
y

n
2, q
y,
s
r
n
s
for s fs
T
y
.
(e)
n

n+1
,
n
is closed under taking the initial segments and the
empty sequence (which too is an initial segment of a play) belongs to

0
.
(f) For any y
n
and T, h either for some z
n+1
, n
z
= n
y
+ 1,
y = z(n
y
+ 1), T
z
= T and h
z
= h or player I has no legal (n
y
+ 1)
th
move p
n
, h
n
(after y was played) such that T
p
n = T, h
n
= h, and
p
n
s
= r
n
s
for s fs
T
(or always or always ).
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


19
There is no problem to carry the denition. Now r
n
s
: n < ) dene a
function d

: if
1
, . . . ,
k

m
2 are distinct then d

(
1
, . . . ,
k
)) = c i for
every (equivalently some)
1
< <
k
from B,

F(

) and r
k
{
1
,...,
k
}

d

k
(
1
, . . . ,
k
) = c.
Now apply 2.7(2) to this coloring, get T

>
2 as there. Now player
I could have chosen initial segments of this T

(in the n
th
move in
n
) and
we get easily a contradiction.
H Stage: We x a winning strategy for player I (whose existence is guar-
anteed by stage G).
We dene a forcing notion Q

. We have (r, y, f) Q

i
(i) r P
cn
a

(ii) y = p

, h

, q

: m(y)) is an initial segment of a play of Gm in which

player I uses his winning strategy
(iii) f is a nite function from B to 0, 1 such that f
1
(1) fs
T
y
(where
T
y
= T
q
m(y) ).
(iv) r = q
y,m(y)
f
1
({1})
.
The Order is the natural one.
I Stage: If J P
cn
a

is dense open then (r, y, f) Q

: r J is dense in
Q

.
Proof. By 3.8(1) (by the appropriate renaming).
J Stage: We dene Q

in V
P

as (r, y, f) Q

: r G

, the order is
as in Q

.
The main point left is to prove the Knaster condition for the partial
ordered set

Q

=

QP

, Q

, a

, e

) demanded in the denition of K

1
. This
will follow by 3.8(3) (after you choose meaning and renamings) as done in
stages K,L below.
K Stage: So let i < , cf(i) ,=
1
, and we shall prove that P
+
+1
/P
i
satises
the Knaster condition. Let p

+1
for <
1
, and we should nd
p P
i
, p
P
i
there is an unbounded A : p

i G

P
i
such that for
any , A, p

, p

are compatible in P

+1
/ G

P
i
.
Without loss of generality:
(a) p

P
cn
+1
.
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


20
(b) for some i

: <
1
) increasing continuous with limit we have:
i
0
> i, cf i

,=
1
, p

P
i
+1
, p

P
i
0
.
Let p
0

= p

i
0
, p
1

= p

= p

i
+1
, p

() = (r

, y

, f

), so without
loss of generality
(c) r

P
i
+1
, r

P
i
0
, m(y

) = m

,
(d) Domf

i
0
[i

, i
+1
),
(e) f

i
0
is constant (remember otp(B) =
1
,
(f) if Domf

= j

0
, . . . j

1
then k

= k, [j

< i

< k

],
_
<k

= j

, f(j

) = f(j

), F(j

))m(y

) = F(j

)m(y

).
The main problem is the compatibility of the q
y

,m(y

)
. Now by the
denition

(in stage E) and 3.8(3) this holds.

L Stage: If c +1 is closed for

Q

, then P

+1
/P
cn
c
satises the Knaster
condition.
If c is bounded in , choose a successor i (supc, ) for

Qi K
1
. We
know that P
i
/P
cn
c
satises the Knaster condition and by stage K, P

+1
/P
i
also satises the Knaster condition; as it is preserved by composition we
have nished the stage.
So assume c is unbounded in and it is easy too. So as seen in stage J,
we have nished the proof of 3.1.
Theorem 3.11. If

, P is the forcing notion of adding Cohen reals

then
()
1
in V
P
, if n < d : []
n
, <
0
, then for some c.c.c. forcing
notion Q we have
Q
there are an uncountable A and an one-
to-one F : A

2 such that d is F-canonical on A (see notation in

2).
()
2
if in V ,
wsp
()

0
(see [Sh289]) and in V
P
, d : []
n
,
<
0
then in V
P
for some c.c.c. forcing notion Q we have
Q
there
are A []

and one-to-one F : A

2 such that d is F-canonical

on A (see 2, ).
()
3
if in V ,
wsp
(
1
)
n

2
and in V
P
d : []
n
, <
0
then in
V
P
for every <
1
and F :

2 for some A of order type

and F

: A

2, F

()
def
= F(otp(A )), d is F

-canonical on A.
()
4
in V
P
, 2

0
(, n)
3
for every <
1
, n < . Really, assuming V [=
GCH, we have
n
1
3
(, n) see [Sh289].
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


21
Proof. Similar to the proof of 3.1. Supercially we need more indiscerni-
bility then we get, but getting M
u
: u [B]
n
) we ignore d(, ) when
there is no u with , M
u
.
Theorem 3.12. If is strongly inaccessible -Mahlo, < , then for some
c.c.c. forcing notion P of cardinality , V
P
satises
(a) MA

(b) 2

0
= = 2

for <
(c) [
1
]
n
,h(n)
for n < , <
0
, h(n) is as in 3.1.
Proof. Again, like 3.1.
4. Partition theorem for trees on large cardinals
Lemma 4.1 Suppose > +
0
and
()

for every -complete forcing notion P, in V

P
, is measurable.
Then
(1) for n < , Pr
f
eht
(, n, ).
(2) Pr
f
eht
(, <
0
, ), if there is > ,
_

+
_
<
2
.
(3) In both cases we can have the Pr
f
ehtn
version, and even choose the
<

: < ) in any of the following ways.

(a) We are given <
0

: < ), and we let for ,

2 T, SP(T)
(T is the subtree we consider):
<

if and only if clp

T
() <
0

clp
T
() where = otp(SP(T))
and clp
T
() = (j) : j lg(), j SP(T)).
(b) We are given <
0

: < ), we let that for ,

2T, SP(T):
<

if and only if n( + 1) <

0
+1
( + 1) where = sup( SP(T)).
Remark. 1) ()

holds for a supercompact after Laver treatment. On

hypermeasurable see Gitik Shelah [GiSh344].
2) We can in ()

restrict ourselves to the forcing notion P actually used.

For it by Gitik [Gi] much smaller large cardinals suce.
3) The proof of 4.1 is a generalization of a proof of Harrington to Halpern
Lauchli theorem from 1978.
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


22
Conclusion 4.2. In 4.1 we can get Pr
f
ht
(, n, ) (even with (3)).
Proof of 4.2. We do the parallel to 4.1(1). By ()

, is weakly compact
hence by 2.6(2) it is enough to prove Pr
f
aht
(, n, ). This follows from 4.1(1)
by 2.6(1).
Proof of Lemma 4.1. 1), 2). Let , (n) < , d
n
Col
n
(n)
(
>
2) for
n < .
Choose such that (
+
)
<2
2
(there is such a by assumption
for (2) and by < for (1)). Let Q be the forcing notion (
>
2, ), and
P = P

be f : dom(f) is a subset of of cardinality < , f(i) Q

ordered naturally. For i , dom(f), take f(i) =<>; Let

i
be the P-name
for f(i) : f G

P
. Let D

be a P-name of a normal ultralter over (in

V
P
). For each n < , d Col
n
(n)
(
>
2), j < (n) and u =
0
, . . . ,
n1
,
where
0
< <
n1
< , let A

j
d
(u) be the P

-name of the set

A
j
d
(u) =
_
i < :

i : < n) are pairwise distinct and

j = d(

0
i, . . . ,

n1
i)
_
.
So A

j
d
(u) is a P

-name of a subset of , and for j(1) < j(2) < (n) we have

j(1)
d
(u) A

j(2)
d
(u) = , and

j<(n)
A

j
d
(u) is a co-bounded subset of
. As
P
D is -complete uniform ultralter on , in V
P
there is exactly
one j < (n) with A
j
d
(u) D. Let j

d
(u) be the P-name of this j.
Let I
d
(u) P be a maximal antichain of P, each member of I
d
(u)
forces a value to j

d
(u). Let W
d
(u) =

dom(p) : p I
d
(u) and W(u) =

W
d
n
(u) : n < . So W
d
(u) is a subset of of cardinaltiy as well as
W(u) (as P satises the
+
-c.c. and p P [ dom(p)[ < ).
As (
++
)
<2
2
, d
n
Col
n

n
(
>
2) there is a subset Z of of
cardinality
++
and set W
+
(u) for each u [Z]
<
such that:
(i) W
+
(u
1
) W
+
(u
2
) = W
+
(u
1
u
2
),
(ii) W(u) W
+
(u) if u [Z]
<
,
(iii) if [u
1
[ = [u
2
[ < and u
1
, u
2
Z then W
+
(u
1
) and W
+
(u
2
) have the
same order type and note that H[u
1
, u
2
]
def
= H
OP
W
+
(u
1
),W
+
(u
2
)
, induces
naturally a map from P u
1
def
= p P : dom(p) W
+
(u
1
) to
Pu
2
def
= p P : dom(p) W
+
(u
2
).
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


23
(iv) if u
1
, u
2
[Z]
<
, [u
1
[ = [u
2
[ then H[u
1
, u
2
] maps I
d
n
(u
1
) onto I
d
n
(u
2
)
and: q j

d
(u
1
) = j H[u
1
, u
2
](q) j

d
(u
2
) = j ,
(v) if u
1
u
2
[Z]
<
, u
3
u
4
[Z]
<
, [u
4
[ = [u
2
[, H
OP
u
2
,u
4
maps u
1
onto
u
3
then H[u
1
, u
3
] H[u
2
, u
4
].
Let (i) be the i
th
member of Z.
Let s(m) be the set of the rst m members of Z and R
n
= p P :
dom(p) W
+
(s(n))

ts(n)
W
+
(t).
We dene by induction on < a function F

and p
u
R
|u|
for
u

<
[

2]
<
where we let

be the empty subset of [

2] and we behave
as if [ ,=

,=

] and we also dene () < , such that:

(i) F

is a function from
>
2 into
>
2, extending F

for < ,
(ii) F

maps

2 to
()
2 for some () < and
1
<
2
< (
1
) <
(
2
),
(iii)
>
2 implies F

() F

(),
(iv) for

2, + 1 < and < 2 wehaveF

()) F

()),
(v) p
u
R
m
whenever u [

2]
m
, m < , < and for u(1) [Z]
m
let
p
u,u(1)
= H[s([u[), u(1)](p
u
) .
(vi)

2, < , then p
{}
(min Z) = F

().
(vii) if < , u [

2]
n
, n < , h : u s(n) one-to-one onto (not necessarily
order preserving) then for some c(u, h) < (n):
_
tu
p
t,h

(t)

P

n
(

(0)
, . . . ,

(n1)
) = c(u, h),
(Note: as p
u
R
|u|
the domains of the conditions in this union are
pairwise disjoint.)
(viii) If n, u, , h are as in (vii), u =
0
, . . . ,
n1
,

2, <
then d
n
(F

(
0
), . . . , F

(
n1
)) = c(u, h) where h is the unique function
from u onto s(n) such that [h(

) h(
m
)

<

m
].
(ix) if < < ,
1
, . . . ,
n1

2, n < , and
0
, . . . ,
n1
are
pairwise distinct then:
p
{
0
,...,
n
}
p
{
0
,...,
n1
}
.
For limit: no problem.
For + 1, limit: we try to dene F

() for

2 such that

<
F
+1
(
) F

() and (viii) holds. Let =

<
(), and for

2, F
0

() =
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


24

<
F

() and for u [

2]
<
, p
0
u
def
=

p
0
{:u}
: < , [u[ = [
: u[. Clearly p
0
u
R
|u|
.
Then let h :

2 Z be one-to-one, such that <

h() < h() and

let p
def
=

p
0
u,u(1)
: u(1) [Z]
<
, u [

2]
<
, [u(1)[ = [u[, h

(u) = u(1).
For any generic G P

to which p belongs, < and ordinals

i
0
< < i
n1
from Z such that h
1
(i

) : < n) are pairwise distinct

we have that
B
{i

:<n},
=
_
< : d
n
(
i
0
, . . . ,
i
n1
) = c(u, h

)
_
,
belongs to D[G], where u = h
1
(i

) : < n and h

: u s([u[) is
dened by h

(h
1
(i

) ) = H
OP
{i

:<n},s(n)
(i

). Really every large enough

< can serve so we omit it. As D[G] is -complete uniform ultralter on
, we can nd (, ) such that B
u
for every u [

2]
n
, n < . We
let for

2, F

() =

h(i)
[G], and we let p
u
= p
0
u
except when u = ,
then:
p
u
(i) =
_
p
0
u
(i) i ,= (0)
F
+1
() i = (0)
.
For + 1, is a successor: First for
1
2 dene F()) = F

()).
Next we let (u
i
, h
i
) : i < i

, list all pairs (u, h), u [

2]
n
, h : u s([u[),
one-to-one onto. Now, we dene by induction on i i

, p
i
u
(u [

2]
<
)
such that :
(a) p
i
u
R
|u|
,
(b) p
i
u
increases with i,
(c) for i + 1, (vii) holds for (u
i
, h
i
),
(d) if
m

2 for m < n, n < ,

m
(1) : m < n) are pairwise distinct,
then p
{
m
(1) :m<n}
p
0
{
m
:m<n}
,
(e) if

2, ( 1) = then p
0
{}
(0) = F

(( 1))).
There is no problem to carry the induction.
Now F
+1


2 is to be dened as in the second case, starting with
p
i

{}
().
For = 0, 1: Left to the reader.
So we have nished the induction hence the proof of 4.1(1), (2).
3) Left to the reader ( the only inuence is the choice of h in stage of the
induction).
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


25
5. Somewhat complimentary negative partition relation in
ZFC
The negative results here suce to show that the value we have for 2

0
in
3 is reasonable. In particular the Galvin conjecture is wrong and that for
every n < for some m < ,
n
, [
1
]
m

0
.
See Erdos Hajnal Mate Rado [EHMR] for
Fact 5.1. If 2
<
< 2

, , []
n

then , [(2
<
)
+
]
n+1

.
This shows that if e.g. in 1.4 we want to increase the exponents, to 3
(and still =
<
) e.g. cannot be successor (when
0
) (by [Sh276],
3.5(2)).
Denition 5.2. Pr
np
(, , ), where =
n
: n < ), means that
there are functions F
n
: []
n

n
such that for every W []

for
some n, F

n
([W]
n
) = (n). The negation of this property is denoted by
NPr
np
(, , ).
If
n
= we write instead of
n
: n < ).
Remark 5.2A. 1) Note that []
<

means: if F : []
<
then for
some A []

, F

([A]
<
) ,= . So for =
0
, , []
<

, (use
F : F() = [[) and Pr
np
(, , ) is stronger than , []
<

.
2) We do not write down the monotonicity properties of Pr
np
they are
obvious.
Claim 5.3 1) We can (in 5.2) w.l.o.g. use F
n,m
: []
n

n
for n, m <
and obvious monotonicity properties holds, and n.
2) Suppose NPr
np
(, , ) and , []
n

or even , []
<

. Then the
following case of Chang conjecture holds:
(*) for every model M with universe and countable vocabulary, there is
an elementary submodel N of M of cardinality ,
[N [ <
3) If NPr
np
(,
1
,
0
) then (,
1
) (
1
,
0
).
Proof. Easy.
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


26
Theorem 5.4. Suppose Pr
np
(
0
, ,
0
), regular >
0
and
1

0
, and
no

(
0
,
1
) is

-Mahlo. Then Pr
np
(
1
, ,
0
).
Proof. Let =
8
(
1
)
+
, let F
0
n,m
: m < list the denable n-
place functions in the model (H(), , <

), with
0
, ,
1
as parameters,
let F
1
n,m
(
0
, . . . ,
n1
) (for
0
, . . . ,
n1
<
1
) be F
0
n,m
(
0
, . . . ,
n1
) if
it is an ordinal <
1
and zero otherwise. Let F
n,m
(
0
, . . . ,
n1
) (for

0
, . . . ,
n1
<
1
) be F
0
n,m
(
0
, . . . ,
n1
) if it is an ordinal < and zero
otherwise. We shall show that F
n,m
(n, m < ) exemplify Pr
np
(
1
, ,
0
)
(see 5.3(1)).
So suppose W [
1
]

is a counterexample to Pr(
1
, ,
0
) i.e. for no
n, m, F

n,m
([W]
n
) = . Let W

be the closure of W under F

1
n,m
(n, m < ).
Let N be the Skolem Hull of W in (H(), , <

), so clearly N
1
= W

.
Note W

1
, [W

[ = . Also as cf() >

0
if A W

, [A[ = then for

some n, m < and u
i
[W]
n
(for i < ), F
1
n,m
(u
i
) A and [i < j <
F
1
n,m
(u
i
) ,= F
1
n,m
(u
i
)]. It is easy to check that also W
1
= F
1
n,m
(u
i
) : i <
is a counterexample to Pr(
1
, , ). In particular, for n, m < , W
n,m
=
F
1
n,m
(u) : u [W]
n
is a counterexample if it has power . W.l.o.g. W is a
counterexample with minimal
def
= sup(W) = +1 : W. The above
discussion shows that [W

[ < for < . Obviously cf =

+
. Let

i
: i < ) be a strictly increasing sequence of members of W

, converging
to , such that for limit i we have
i
= min(W

j<i
(
j
+ 1). Let
N =

i<
N
i
, N
i
N, [N
i
[ < , N
i
increasing continuous and w.l.o.g.
N
i
= N
i
.
Fact: is >
0
.
Proof. Otherwise we then get an easy contradiction to Pr(
0
, , )) as
choosing the F
0
n,m
we allowed
0
as a parameter.
Fact: If F is a unary function denable in N, F() is a club of for every
limit ordinal (<
1
) then for some club C of we have
(j C min C)(i
1
< j)(i (i
1
, j))[i C
i
F(
j
)].
Proof. For some club C
0
of we have j C
0
(N
j
,
i
: i < j, W)
(N,
i
: i < , W).
We let C = C

0
= acc(C) (= set of accumulation points of C
0
).
We check C is as required; suppose j is a counterexample. So j =
sup(j C) (otherwise choose i
1
= max(j C)). So we can dene, by
induction on n, i
n
, such that:
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


27
(a) i
n
< i
n+1
< j
(b)
i
n
, F(
j
)
(c) (
i
n
,
i
n+1
) F(
j
) ,= .
Why (C

0
)? [= F(
j
) is unbounded below
j
hence N [= F(
j
) is
unbounded below
j
, but in N,
i
: i C
0
, i < j is unbounded below

j
.
Clearly for some n, m,
j
W
n,m
(see above). Now we can repeat the
proof of [Sh276,3.3(2)] (see mainly the end) using only members of W
n,m
.
Note: here we use the number of colors being
0
.

+
Fact: Wolog the C in Fact is .
Proof: Renaming.
Fact: is a limit cardinal.
Proof: Suppose not. Now cannot be a successor cardinal (as cf =

0
< ) hence for every large enough i, [
i
[ = [[, so [[ W

N and
[[
+
W

.
So W

[[ has cardinality < hence order-type some

< . Choose
i

< limit such that [j < i

j +

< i

]. There is a denable function

F of (H(), , <

) such that for every limit ordinal , F() is a club of ,

0 F(), if [[ < , F() [[ = , otp(F()) = cf .
So in N there is a closed unbounded subset C

j
= F(
j
) of
j
of order
type cf
j
[[, hence C

j
N has order type

, hence for i

chosen
above unboundedly many i < i

,
i
, C

. We can nish by fact

+
.
Fact: For each i < ,
i
is a cardinal.
Proof: If [
i
[ < i then [
i
[ N
i
, but then [
i
[
+
N
i
contradicting to Fact
, by which [
i
[
+
< , as we have assumed N
i
= N
i
.
Fact: For a club of i < ,
i
is a regular cardinal.
(Proof: if S = i :
i
singular is stationary, then the function
i
cf(
i
)
is regressive on S. By Fodor lemma, for some

< , i < : cf
i
<

is
stationary. As [N

[ < for some

, i < : cf
i
=

is stationary.
Let F
1,m
() be a club of of order type cf(), and by fact we get a
contradiction as in fact .
Fact: For a club of i < ,
i
is Mahlo.
Proof: Use F
1,m
() = a club of which, if is a successor cardinal or
inaccessible not Mahlo, then it contains no inaccessible, and continue as in
fact .
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


28
Fact: For a club of i < ,
i
is
i
-Mahlo.
Proof: Let F
1,m(0)
() = sup : is -Mahlo. If the set i < :
i
is not

i
-Mahlo is stationary then as before for some N, i : F
1,m(0)
(
i
) =
is stationary and let F
1,m(1)
() a club of such that if is not ( + 1)-
Mahlo then the club has no -Mahlo member. Finish as in the proof of fact
.
Remark 5.4.A. We can continue and say more.
Lemma 5.5 1) Suppose > > are regular cardinals, n 2 and
(i) for every regular cardinal , if > then , []
<
(1)
.
(ii) for some () < for every regular ((), ), , [()]
n
(2)
.
Then
(a) , []
n+1

where = min(1), (2),

(b) there are functions d
2
: []
n+1
(2), d
1
: []
3
(1) such that for
every W []

, d

1
([W]
3
) = (1) or d

2
([W]
n+1
) = (2).
2) Suppose > > are regular cardinals, and
(i) for every regular [, ), , []
<
(1)
,
(ii) sup < : regular , []
n
(2)
.
Then
(a) , []
2n

where = min(1), (2)

(b) there are functions d
1
: []
3
(1), d
2
: []
2n
(2) such that for
every W []

, d

1
([W]
3
) = (1) or d

2
([W]
2n
= (2).
Remark. The proof is similar to that of [Sh276] 3.3,3.2.
Proof. 1) We choose for each i, 0 < i <
i
, C
i
such that: if i is a successor
ordinal, C
i
= i 1, 0; if i is a limit ordinal, C
i
is a club of i of order type
cf i, 0 C
i
, [cf i < i cf i < min(C
i
0)] and C
i
acc(C
i
) contains only
successor ordinals.
Now for < , > 0 we dene by induction on ,
+

(, ),

(, ),
and then (, ), (, ).
(A)
+
0
(, ) = ,

0
(, ) = 0.
(B) if
+

(, ) is dened and > and is not an accumulation point of

C

(,)
then we let

+1
(, ) be the maximal member of C

(,)
which is < and
+
+1
(, ) is the minimal member of C

(,)
which
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


29
is (by the choice of C

(,)
and the demands on
+

(, ) they are
well dened).
So
(B1) (a)

(, ) <
+

(, ), and if the equality holds then

+
+1
(, )
is not dened.
(b)
+
+1
(, ) <
+

(, ) when both are dened.

(C) Let k = k(, ) be the maximal number k such that
+
k
(, ) is dened
(it is well dened as
+

(, ) : < ) is strictly decreasing). So

(C1)
+
k(,)
(, ) = or
+
k(,)
> ,
+
k(,)
is a limit ordinal and is an
accumulation point of C

+
k(,)
(, ).
(D) For m k(, ) let us dene

m
(, ) = max

(, ) + 1 : m.
Note
(D1) (a)
m
(, ) (if dened),
(b) if is limit then
m
(, ) < (if dened),
(c) if
m
(, ) then for every m we have

(, ) =
+

(, ),

(, ) =

(, ),

(, ) =

(, ).
(explanation for (c): if
m
(, ) < this is easy (check the denition)
and if
m
(, ) = , necessarily = and it is trivial).
(d) if m then

(, )
m
(, )
For a regular ((), ) let g
1

: []
<
(2) exemplify , []
<
(1)
and for every regular cardinal [, ) let g
2

: []
n
(2) exemplify
, [()]
n
(2)
. Let us dene the colourings:
Let
0
>
1
> . . . >
n
. Remember n 2.
Let n = n(
0
,
1
,
2
) be the maximal natural number such that:
(i)
n
(
0
,
1
) <
0
is well dened,
(ii) for n,

(
0
,
1
) =

(
0
,
2
).
We dene d
2
(
0
,
1
, . . . ,
n
) as g
2

(
1
, . . . ,
n
) where
=cf (
+
n(
0
,
1
,
2
)
(
0
,
1
)),

=otp
_

+
n(
0
,
1
,
2
)
(
0
,
1
)
_
.
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


30
Next we dene d
1
(
0
,
1
,
2
) .
Let i() = sup
_
C

+
n
(
0
,
2
)
C

+
n
(
1
,
2
)
_
where n = n(
0
,
1
,
2
), E be
the equivalence relation on C

+
n
(
0
,
1
)
i() dened by

1
E
2
C

+
n
(
0
,
2
)
[
1
<
2
< ].
If the set w =
_
C

+
n
(
0
,
1
)
: > i(), = min /E
_
is nite,
we let d
1
(
0
,
1
,
2
) be g
1

: w) where =

+
n
(
0
,
1
)

=
otp
_
C

+
n
(
0
,
1
)
_
.
We have dened d
1
, d
2
required in condition (b) ( though have not yet
proved that they work) We still have to dene d (exemplifying , []
n+1

).
Let n 3, for
0
>
1
> . . . >
n
, we let d(
0
, . . . ,
n
) be d
1
(
0
,
1
,
2
) if w
dened during the denition has odd number of members and d
2
(
0
, . . . ,
n
)
otherwise.
Now suppose Y is a subset of of order type , and let = supY . Let
M be a model with universe and with relations Y and (i, j) : i C
j
. Let
N
i
: i < ) be an increasing continuous sequence of elementary submodels
of M of cardinality < such that (i) =
i
= min(Y N
i
) belongs to N
i+1
,
sup(N
i
) = sup(N ). Let N =

i<
N
i
. Let (i) =
i
def
= sup(N
i

i
),
so 0 <
i

i
, and let n = n
i
be the rst natural number such that
i
an
accumulation point of C
i
def
= C

+
n
(
i
,(i))
, let
i
=
n(i)
(
i
,
i
). Note that

+
n
(
i
,
i
) =
+
n
(
i
,
i
) hence it belongs to N.
Case I: For some (limit) i < , cf(i) and ( < i)[ + () < i] such
that for arbitrarily large j < i, C
i
N
j
is bounded in N
j
= N
j

j
.
This is just like the last part in the proof of [Sh276],3.3 using g
1

and d
1
for
= cf(
+
n
i
(
i
,
i
).
Case II: Not case I.
Let S
0
= i < : ( < i)[ +() < i], cf(i) = . So for every i S
0
for some j(i) < i, (j)
_
j (j(i), i) C
i
N
j
is unbounded in
j
_
. But as
C
i

i
is a club of
i
, clearly (j)
_
j (j(i), i)
j
C
i
_
.
We can also demand j(i) >
n((i),(i))
((i), (i)).
As S
0
is stationary, (by not case I) for some stationary S
1
S
0
and
n(), j() we have (i S
1
)
_
j(i) = j() n((i),
i
) = n()
_
.
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


31
Choose i() S
1
, i() = sup(i() S
1
), such that the order type of
S
1
i() is i() > (). Now if i
2
< i
1
S
1
i() then n(
i()
,
i
1
,
i
2
) =
n(). Now L
i()
def
=
_
otp(
i
C
i()
) : i S
1
i()
_
are pairwise distinct
and are ordinals <
def
= [C
i()
[, and the set has order type (). Now apply
the denitions of d
2
and g
2

on L
i()
.
2) The proof is like the proof of part (1) but for
0
>
1
> we let
d
2
(
0
, . . . ,
2n1
) = g
2

(
0
, . . . ,
n
) where

def
= otp (C

+
n
(
2
,
2+1
)
(
2
,
2+1
)
2+1
)
and in case II note that the analysis gives possible

s so that we can
apply the denition of g
2

.
Denition 5.7. Let ,
stg
[]
n

mean: if d : []
n
, and
i
: i < ) is
strictly increasingly continuous and for i < j < ,
i,j
[
i
,
i+1
) then
=
_
d(w) : for some j < , w [
i,j
: i < j]
n
_
.
Lemma 5.8. 1)
t
, [
1
]
n+1

0
for n 1.
2)
n
,
stg
[
1
]
n+1

0
for n 1.
Proof. 1) For n = 2 this is a theorem of Torodcevic, and if it holds for
n 2 by 5.5(1) we get that it holds for n+1 (with n, , , , (), (1),
(2) there corresponding to n + 1,
n+1
,
1
,
0
,
0
,
0
,
0
here).
2) Similar.
References
[EHMR] P. Erd os, A. Hajnal, A. Mate, R. Rado, Combinatorial set theory:
partition relations for cardinals, Disq. Math. Hung., no 13, (1984)
[Gi] M. Gitik, Measurability preserved by -complete forcing notion
[GiSh344] M. Gitik, S. Shelah, On certain indiscernibility of strong cardinals and
a question of Hajnal, Archive f ur Math. Logic, 28(1989) 3542
[HL] Halpern, Lauchli,
2
8
8


r
e
v
i
s
i
o
n
:
1
9
9
3
-
0
8
-
2
7







m
o
d
i
f
i
e
d
:
1
9
9
3
-
0
8
-
2
7


32
[Sh80] S. Shelah, A Weak Generalization of MA to Higher Cardinals, Israel J.
of Math, [BSF], 30(1978), 297-306.
[Sh289] S. Shelah, Consisting of Partition Theorem for Graphs and Models,
Proc. of the Torontp 8/87 Conference in General Topology.
[Sh276] S. Shelah, Was Sierpinsky right I ? Israel J. Math. 62 (1988) 355380
[ShSt154] S. Shelah and L. Stanley, Generalized Martin Axiom and the Souslin
Hypothesis for Higher Cardinals, Israel J. of Math, [BSF], 43(1982),
225-236.
[ShSt154a] S. Shelah and L. Stanley, Corrigendum to Generalized Martins Ax-
iom and Souslins Hypothesis for Higher Cardinal, Israel J. of Math,
53(1986), 309-314.
[To] S. Todorcevic, Partitions of pairs of uncountable ordinals, Acta MAth,
159 (l987), 261294
Saharon Shelah
Institute of Mathematics
The Hebrew University
Jerusalem, Israel