(

7
6
1
)


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







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


A PARTITION RELATION USING
STRONGLY COMPACT CARDINALS
SH761
Saharon Shelah
Institute of Mathematics
The Hebrew University
Jerusalem, Israel
Rutgers University
Mathematics Department
New Brunswick, NJ USA
Abstract. If is strongly compact and > and is regular (or alternatively
cf() ), then
`
2
<

+
( +)
2

holds for , < .

2000 Mathematics Subject Classication. 2000 Math Subject Classication: 03E02.
I would like to thank Alice Leonhardt for the beautiful typing.
Research of the author was partially supported by the United States-Israel Binational Science
Foundation.
Publ.761.
Latest Revision - 02/Oct/21
Typeset by A
M
S-T
E
X
1
(
7
6
1
)


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







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


2 SAHARON SHELAH
0 Introduction
The aim of this paper is to prove the following theorem.
0.1 Theorem. If is a strongly compact cardinal, > is regular and , <
then the partition relation

2
<

+
( +)
2

holds.
0.2 Theorem. In Theorem 0.1.
Instead regular, cf() > suces.
We notice that our argument is valid in the case = . As for the history
of the problem we point out that Hajnal proved in an unpublished work, that
(2

)
+
(
1
+ n)
2
2
holds for every n < . Then it was showed in [Sh 26, 6] that
for > regular and 2
||
< , the relation (2
<
)
+
(+)
2
2
is true. More recently
Baumgartner, Hajnal, and Todorcevic in [BHT93] extended this to the case when
the number of colors is arbitrary nite. Earlier by [Sh 424], we have (2
<
)
+n

( m)
2
k
for n large enough (this was complimentary to the main result there that

0
< =
<
+ 2

arbitrarily large does not imply 2

( )
2
2
). Subsequently
[BHT93] improves n. We hope that the way the strong compactness was used will
be useful elsewhere; see [Sh 666] for a discussion of a possible consistency of failure.
I also thank Peter Komjath for improving the presentation.
Notation. If S is a set, a cardinal then [S]

= a S : [a[ = , [S]
<
= a
S : [a[ < . If D is some lter over a set S then X D
+
denotes that S X / D
and X S. If < are regular cardinals then S

= < : cf() = , a
stationary set. The notation A = x

: <
<
, etc., means that A is enumerated
increasingly.
(
7
6
1
)


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







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


A PARTITION RELATION USING STRONGLY COMPACT CARDINALS SH761 3
1 The case of regular
1.1 Lemma. Assume =

. Assume that D is a normal lter on

+
and
A

D
+
satises A

cf() > , and F

is a function with domain [A

]
2
and range of cardinality . Then there are a normal lter D
0
on
+
extending
D, A
0
D
0
with A
0
A

and C
0
Rang(F

) satisfying Rang(F

[A
0
]
2
) = C
0
such that: if X D
+
0
then Rang(F

[X]
2
) C
0
.
We rst prove a claim
1.2 Claim. Assume =

and F

: [S

]
2
C

, [C

[ , D is a normal lter on

+
, S

+
belongs to D
+
and S

cf() > . There is a set A D

+
such
that A S

and some C C
0
satisfying Rang(F

[A]
2
) = C and: if f : A
+
is a regressive function, then for some <
+
we have Rang(F

[f
1
()]
2
) = C
and f
1
() is a subset of
+
from D
+
.
Proof. Toward contradiction assume that no such sets A, C exist. We build a tree
T as follows. Every node t of the tree will be of the form
t = A

: ), f

: < ), i

: < ))
=

A
t

: ), f
t

: < ), i
t

: < )

for some ordinal = (t) where A

: ) is a decreasing, continuous sequence

of subsets of
+
; for every < , f

is a regressive function on A

; and i

: < )
is a sequence of distinct elements of C

. It will always be true that if t <

T
t

, then
each of the three sequences of t

extend the corresponding one of t.

To start, we make the node t with (t) = 0, A
0
= S

the root of the tree.

At limit levels we extend (the obvious way) all conal branches to a node.
If we are given an element t = A

: ), f

: < ), i

: < )) of the
tree and the set A

is = mod D then we leave t as a terminal node. Otherwise,

let C = C
t
= Rang

[A

]
2

and notice that by hypothesis, toward contradic-

tion, the pair A

, C
t
cannot be as required in the Claim. There is, therefore, a
regressive function f = f
t
with domain A

, such that for every x <

+
the set
Rang

[f
1
(x)]
2

is a proper subset of C
t
or f
1
(x) is a = mod D subset
of
+
. We make the immediate extensions of t the sequences of the form a t
x
=
A

: + 1), f

: < + 1), i

: < + 1)) where A

+1
= f
1
(x), f

= f
t
and i

C
t
is some colour value such that: if A
+1
,= mod D then i

is not in
the range of F

[A

]
2
.
(
7
6
1
)


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







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


4 SAHARON SHELAH
Having constructed the tree observe that every element x S

+
belongs
to a set A
t(x)
(x)
for some (unique) terminal node t(x) of T. Also, (x) <
+
(<
+
)
holds by the selection of the i

s as i
t(x)

: < (x)) is a sequence of members of

C

with no repetitions while C

, the set of colours, has members. For some

set S S

of ordinals x <
+
which belong to D
+
(by the normality of D) the
value of (x) is the same, say . For x S we let g

(x) = f
t(x)

(x) where f
t(x)

is
the -th regressive function in the node t(x) T. Again, by

= & (
S)[cf() > ] we have that (x S

)( < )g

(x) =

holds for some sequence

: < ) and subset S

S from D
+
. But then we get that the set S

satises
x, y S

(A
t(x)

, f
t(x)

, i
t(x)

) = (A
t(y)

, f
t(y)

, i
t(y)

) for every < ; we can prove

this by induction on . We can then prove that A
t(x)

= A
t(y)

for x, y S

. We can
conclude that x, y S

t(x) = t(y), so S

A
t
(t)
for some terminal node t, but
this latter set is in D
+
, a contradiction.

1.2
Proof of Lemma 1.1. Apply Claim 1.2 with S

= A

to get corresponding (C, A).

Dene the ideal I as follows. For X
+
we let X I i there are a member E of
D and a regressive function f : X A
+
such that every Rang

[f
1
()]
2

is a proper subset of C or f
1
() is a = modD subset of
+
.
Now:
1.3 Claim. I is a normal ideal on
+
(and A

=
+
mod I).
Proof. Straightforward.
Set D
0
to be the dual lter of I, let A
0
= A and let C
0
= C; by 1.2 we are
done.
1.1
1.4 Remark. 1) If Lemma 1.1 holds for some D
0
, A
0
, C
0
then it holds for D
1
, A
1
, C
0
when the normal lter D
1
extends D
0
, and A
1
D
1
satises A
1
A
0
.
2) If D
0
, A
0
, C
0
satisfy Lemma 1.5, and X D
+
0
then X contains a homogeneous
set of order type + 1 of color for every C
0
.
3) Lemma 1.1 is closely related to the proof in [Sh 26].
Proof of Theorem 0.1. Let = 2
<
, and F : [
+
]
2
be a colouring; we apply
1.1 for A

= S

+
cf()
, (F = F, = , = ) and D the club lter. We shall write
F(, ) for F(, ) and 0 for F(, ).
We x A
0
, D
0
, C
0
which we get by 1.1.
(
7
6
1
)


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







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


A PARTITION RELATION USING STRONGLY COMPACT CARDINALS SH761 5
1.5 Lemma. Almost every A
0
; (i.e. for all but a set = mod D
0
) satises
the following: if s [A
0
]
<
and z

: <
<
A
0
[,
+
) with < then
there is y

: <
<
A
0
(sup(s), ) such that:
(a) F(x, y

) = F(x, z

) (for x s, < );
(b) F(y

, y

) = F(z

, z

) (for < < ).

Proof. By simple reection (using the regularity of ).
1.6 Lemma. There
1
is A

0
A
0
, A

0
D
0
such that: if A

0
, s []
<
and
C
0
, then there exists a
1
A
0
, <
1
such that
(a) F(x, ) = F(x,
1
) (for x s);
(b) F(,
1
) = .
Proof. Otherwise, there is some X A
0
, X D
+
0
such that for every X there
are s() []
<
and () C
0
such that there is no
1
> satisfying (a) and (b).
By normality and =
<
we can assume that s() = s and () = holds for
X. By Lemma 1.1, that is the choice of (A
0
, D
0
, C
0
), there must exist <
1
in X with F(,
1
) = and this is a contradiction.

1.6
Continuation of the proof of Theorem 0.1. Let A

0
A
0
satisfy Lemmas 1.1 and
1.6 and pick some
1
A

0
and then let T = A

0
(
1
+ 1).
1.7 Lemma. There exists a function G : T T C
0
such that: if s [
1
]
<
, <
, and Z = z

: <
<
T then there is y

: <
<
(sup(s),
1
) such that
(a) F(x, y

) = F(x, z

) (for x s, < );
(b) F(y

, y

) = F(z

, z

) (for < < );

(c) F(y

, z

) = G(z

, z

) (for , < ).
1
in fact, if A

1
D
+
0
then for some A

0
A
1
A
0
, A
1
\A

0
= modulo D
0
and the conclusion
holds for every A

0
(
7
6
1
)


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







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


6 SAHARON SHELAH
Proof. As is strongly compact, it suces to show that for every Z [T]
<
there
exists a function G : Z Z as required. Clauses (a) and (b) are obvious by
Lemma 1.5, and it is clear that, if we x Z, then for every s [
1
]
<
there is an
appropriate G : Z Z . We show that there is some G : Z Z that works
for every s. Assume otherwise, that is, for every G : Z Z there is some
s
G
[
1
]
<
such that G is not appropriate for s
G
. Notice that the number of these
functions G is less than . Then no G could be right for s = s
G
: G a function
from Z Z to [
1
]
<
, a contradiction.
1.7
Continuation of the proof of Theorem 0.1. We now apply Lemma 1.1 to the colour-
ing

Gx, y =

G(x, y) = F(x, y), G(x, y)) for x < y in T and 0 otherwise, and the
lter D
0
and the set T and get the normal lter D
1
D
0
, the set A
1
T A

0
such
that A
1
D
1
and the colour set C
1
. Notice that actually C
1
C
0
C
0
.
We can also apply Lemmas 1.5 and 1.6 and get some set A

1
A
1
.
1.8 Lemma. There is a set a [A

1
]
<
such that for every decomposition a =
a

:

C
1
there is some

C
1
such that
() for every C
1
there is an -homogeneous subset for the colouring

G of
order type in a

,
() similarly for every C
0
and F.
Proof. This follows from the strong compactness of as A

1
itself has this partition
property (or more details in 2.8).
1.8
Continuation of the Proof of 0.1. Fix a set a as in 1.8.
We now describe the construction of the required homogeneous subset. Let

2
A

1
be some element with
2
> sup(a). For

= (
1
,
2
) C
1
let a

be
the following set:
a

= x a :

G(x,
2
) =

.
By Lemma 1.8, there is some

= (
1
,
2
) C
1
for which the statement in 1.8 above
is true and necessarily (as a
2
A

1
A
0
and a

,= ) we have
1
,
2
C
0
.
Select some b a

, otp(b) = such that F is constantly

2
on b; this is possible by
clause () of 1.8. This set b will be the part of our homogeneous set of ordinals
of order type +, so we will have to construct a set of order type below b. By
(
7
6
1
)


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







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


A PARTITION RELATION USING STRONGLY COMPACT CARDINALS SH761 7
induction on we will choose x

such that the set x

: <
<

1
satises the
following conditions:
()
1
F(x

, x

) =
2
(for < ),
()
2
F(x

, b
2
) =
2
, i.e. F(x

, y) =
2
when y b
2
.
Assume that we have reached step , that is, we are given the set of ordinals with
x

: <
<
and call this set s. Applying Lemma 1.6 for A
1
, A

1
,
2
and s b
and the colouring

G here standing for A
0
, A

0
, , s and the colouring F there (that
is the choice of A

1
) we get that there exists some
3
>
2
(standing for
1
there)
such that
(i)
3
A
1
(ii)

G(x,
3
) =

G(x,
2
) for x s b
(iii)

G(
2
,
3
) = (
1
,
2
),
hence:
()
3
F(x

,
3
) =
2
(for < ).
[Why? As F(x

,
3
) = F(x

,
2
) by (ii) and the choice of

G and F(x

,
2
) =

2
by ()
2
from the induction hypothesis.]
()
4
G(b
2
,
3
) =
2
, i.e. G(y,
3
) =
2
when y b
2
.
[Why? If y b then by (ii) and the denition of

G we have G(y,
3
) =
G(y,
2
), but b a

so by the choice of a

we have G(y,
2
) =
2
. For
y =
2
use clause (iii) that is (
1
,
2
) =

G(
2
,
3
) = (F(
2
,
3
), G(
2
,
3
)).]
By the choice of G this implies that there is some x

as required; that is by the

choice of

G (see Lemma 1.7), applied to Z = z
i
: i < enumerating the set
b
2
,
3
and s as above, we get y
i
: i < , now necessarily
3
= z
1
, and we
can choose y
1
as x

.
1.1
(
7
6
1
)


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







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


8 SAHARON SHELAH
2 The case of singular
We prove version 0.2 of the main theorem.
Proof of Theorem 0.2. Let = cf(). Let =

<

with

> > strictly

increasing. Let

= 2

and =

: < = 2
<
. We also x F : [
+
]
2
.
2.1 Claim. For some

C we have:
(a)

C = C

: S)
(b) S
+
, C

(c) otp(C

)
(d) S

= < : otp(C

) = is stationary
(e) C

unbounded in if otp(C

) =
(f) C

S & C

= C

2.1
Proof. By [Sh 420, 1] as
+
<
+
, = cf().
Continuation of the proof of 0.2: Let D
0
, A
0
, C
0
be as given by Lemma 1.1 with
the club lter of
+
, S

(from clause (d) of 2.1 above) here standing for D, A

there
so A
0
S

.
Notation: () = otp(C

).
2.2 Claim. Let > 2

, <

a well ordering of H ()). For any x H () we can

nd

B = B

: < ) such that:

(a) B

(H (), , <

)
(b)

, , F,

: < ),

C, A
0
, C
0
, D
0
belong to B

(c) B

: < ) B

if / S

(d) |B

| =
()
and [B

()
B

and
()
+ 1 B

(actually follows)
(e) B

= B

: C

if S

.
Proof. Straightforward.
2.3 Observation. 1) We have () < () and B

and B

if C

.
(
7
6
1
)


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







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


A PARTITION RELATION USING STRONGLY COMPACT CARDINALS SH761 9
2.4 Claim. There is a set A

0
A
0
such that
() A

0
D
0
and < A

0
sup(B

+
) <
() if C
0
and A

0
and s [ B

()
: C

, then there is

1
A
0
such that <
1
and
(a) F(x, ) = F(x,
1
) for x s
(b) F(,
1
) = .
Proof. Requirement () holds for all but a non stationary set of A
0
. Require-
ment () is proved as in 1.6.
2.4
Now x A

0
A
0
as in 2.4, and x
1
A

1
and let T = A

0
(
1
+ 1). Recall

1
A

0
S

= : otp(C

) = , = sup(C

) <
+
: cf() = .
2.5 Claim. There is a function G

: T T C
0
such that:
if s [ B

and = () and C

1
and < and Z = z

: <

<
T, then there is y

: <
<
B

=
+
B

such that:
(a) F(x, y

) = F(x, z

) for x s, <
(b) F(,
1
) =
F(z

1
, y

2
) = G(y

1
, y

2
) for
1
,
2
<
(c) F(z

1
, z

2
) = F(y

1
, y

2
) for
1
,
2
<
(d) y
0
> sup(s).
Proof. Like 1.7.
2.6 Claim. There exists a function G : T T C
0
such that if s [T]
<
, then
for arbitrarily large < we have G (s s) = G

(s s).
Proof. Let D

be a uniform -complete ultralter on and dene G by G(, ) is

the unique C
0
such that < : G

(, ) = D

.
2.6
Continuation of the Proof of 0.2. Now we apply Lemma 1.1 to the colouring

G
where

Gx, y =

G(x, y) = (F(x, y), G(x, y)) for x < y in T and zero otherwise and
the lter D
0
and the set T. We get a normal lter D
1
and a set A
1
T A

0
and
a set of colours C
1
. As A
1
A
0
necessarily C
1
C
0
C
0
.
(
7
6
1
)


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







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


10 SAHARON SHELAH
2.7 Claim. There is A

1
A
1
such that:
() A
1
A

1
= mod D
1
() if A

1
, C

and s [ B

()
and

C
1
, then for some

we
have <

A
1
and
(a)

G(x, ) =

G(x,
1
) for every x s
(b)

G(,

) =

.
Proof. Like the proof of 1.6
2.7
2.8 Claim. There is a set a [A

1
]
<
such that:
for every decomposition of a as a

:

C
1
there is

C
1
such that
() for every C
1
there is b a

of order type such that

G [b]
2
is
constantly
() for every C
0
there is b a

of order type such that F [b]

2
is
constantly .
Proof. The claim holds since A

1
has this property and is strongly compact. If
A

1
= a

:

C
1
for some

, a

D
+
1
hence clause () holds by the choice of
D
1
, C
1
; and clause () holds as D
+
1
D
+
0
(as D
0
D
1
) and the choice of D
0
, C
0
.

2.8
Continuation of the proof of 0.2. Now choose
2
A

1
such that
2
> sup(a) and
for

= (
1
,
2
) C
1
dene a

as
a

= x a :

G(x,
2
) =

.
Clearly a

:

C
1
) is a decomposition of a and so there is

= (
1
,
2
) C
1
as
guaranteed by of 2.8. In particular, there is b a

of order type such that

F [b]
2
is constantly
2
(note that (
1
,
2
) C
1
C
0
C
0
so
2
C
0
). Now let
E = < : G

(,
2
) = G(,
2
) for every b. By the denition of G this is
an unbounded subset of and clearly
() if E and b then G

(,
2
) = G(,
2
) = (
1
,
2
).
(
7
6
1
)


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







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


A PARTITION RELATION USING STRONGLY COMPACT CARDINALS SH761 11
For < let () = Min E : <

and let C

1
= () : <
<
.
Now we try to choose by induction on < a element x

satisfying
()
0
x

<
1
and moreover x

1
B
(())
, and < x

< x

()
1
F(x

, x

) =
2
for <
()
2
F(x

, ) =
2
for b
2
.
At step , by 2.7, that is by the choice of A

1
applying clause () there with
x

: < b,
2
,

here standing for s, ,

there, we can nd
3
satisfying the
requirement there on
1
, so
(i)
2
<
3
A
1
(ii)

G(x,
3
) =

G(x,
2
) for x s b
(iii)

G(
2
,
3
) = (
1
,
2
).
Now
()
3
F(x

,
3
) =
2
for < .
[Why? By (ii) we have

G(x

,
3
) =

G(x

,
2
) hence F(x

,
3
) = F(x

,
2
)
but the latter by ()
2
is equal to
2
.]
()
4
G(,
3
) =
2
for b
[Why? By (ii) and as b

G(,
2
) = (
1
,
2
) G(,
2
) =
2
).]
()
5
G(
2
,
3
) =
2
[Why? By clause (iii).]
()
6
x

: < is a subset of
1
B
(())
.
Let y
i
: i < + 2) list b
2
,
3
increasing order.
Now we use the choice of G
()
to choose an increasing sequence z
i
: i < +2)
in
1
B
(())
, z
0
> x

for < such that F(z

i
, y
j
) = G(y
i
, y
j
) for i, j < + 2
and F(x

, z
i
) = F(x

, y
i
) for i < + 2. Let x

= z
+1
so x

=
1
B
(())
is
> x

for < .
Also x

satises ()
0
of the recursive denition. Now < F(x

, x

) =
F(x

, z
+1
) = F(x

, y
+1
) = F(x

,
3
) which is
2
by ()
3
above, so for our choice
of x

, ()
1
holds. Next if b
2
then F(x

, x

) = F(x

, z
+1
) = G(x

,
3
)
which is
2
by ()
4
or ()
5
. So x

is as required.
0.2
(
7
6
1
)


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







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


12 SAHARON SHELAH
REFERENCES.
[BHT93] James Baumgartner, Andras Hajnal, and Stevo Todorcevic. Extensions
of the ErdosRado Theorems. In Finite and Innite Combinatorics in Set
Theory and Logic, pages 118. Kluwer Academic Publishers, 1993. N.W.
Sauer et. al. eds.
[Sh 26] Saharon Shelah. Notes on combinatorial set theory. Israel Journal of
Mathematics, 14:262277, 1973.
[Sh 420] Saharon Shelah. Advances in Cardinal Arithmetic. In Finite and In-
nite Combinatorics in Sets and Logic, pages 355383. Kluwer Academic
Publishers, 1993. N.W. Sauer et al (eds.). 0708.1979.
[Sh 424] Saharon Shelah. On CH+2

1
()
2
2
for <
2
. In Logic Colloquium90.
ASL Summer Meeting in Helsinki, volume 2 of Lecture Notes in Logic,
pages 281289. Springer Verlag, 1993. math.LO/9308212.
[Sh 666] Saharon Shelah. On what I do not understand (and have some-
thing to say:) Part I. Fundamenta Mathematicae, 166:182, 2000.
math.LO/9906113.