9

0
9


r
e
v
i
s
i
o
n
:
2
0
0
9
-
1
0
-
2
5







m
o
d
i
f
i
e
d
:
2
0
0
9
-
1
0
-
2
5


Uniforming n-place Functions on Well Founded Trees
Esther Gruenhut and Saharon Shelah
Abstract. In this paper the Erd os-Rado theorem is generalized to the class of
well founded trees. We dene an equivalence relation on the class ds()
<
0
( nite sequences of decreasing sequences of ordinals) with
0
equivalence
classes, and for n < a notion of n-end-uniformity for a colouring of ds()
<
0
with colours. We then show that for every ordinal , n < and cardinal
there is an ordinal so that for any colouring c of T = ds()
<
0
with
colours, T contains S isomorphic to ds() so that cS
<
0
is n-end uniform.
For c with domain T
n
this is equivalent to nding S T isomorphic to ds()
so that cS
n
depends only on the equivalence class of the dened relation, so
in particular T (ds())
n
,
0
. We also draw a conclusion on colourings of
n-tuples from a scattered linear order.
0. Introduction
This paper deals with a Ramsey-type theorem for scaterred order types. We
dedicate this section to some general background. A Ramsey-type theorem begins
with a target element and a xed number of colors, . The statement asserts
that there exists another element (of the same type) so that for every coloring of
by colors, one can nd a monochromatic -copy included in .
The simplest example is the class of innite cardinals, and coloring functions dened
on singletons. For instance,
+
(
+
)
1

holds for every innite cardinal . It

means that for any coloring c :
+
there exists a copy of
+
(namely, a subset
of
+
whose cardinality is
+
) which is monochromatic under c.
This simple version works for order types as well. Given any order type (this is
the target), and a xed number of colors , one can nd an order type so that
()
1

(i.e., for every coloring c : there exits a monochromatic copy of

in ).
We concentrate, throughout the paper, in the interesting class of scaterred order
types. Let us start with the following:
Definition 0.1. Scaterred order types.
(1) is the order type of the set of rational numbers (Q, <)
(2) For two order types , we say that i there is an order preserving
embedding of into
Research supported by the United States-Israel Binational Science Foundation (Grant no.
2002323). Publication 909 in Shelahs archive.
AMS 2000 classication 03E05, 05C15.
Key words: Set Theory, partition relation, well founded trees, scattered order types.
1
9
0
9


r
e
v
i
s
i
o
n
:
2
0
0
9
-
1
0
-
2
5







m
o
d
i
f
i
e
d
:
2
0
0
9
-
1
0
-
2
5


2 ESTHER GRUENHUT AND SAHARON SHELAH
(3) An order type is scattered when ( )
The investigation of scaterred order types goes back to Hausdor. This denition
is a negative one. Hausdor proved in [3] that the class of scaterred order types
is characterized by a simple positive closure property. This class is the smallest
class which contains 0, 1 and is closed under well ordered and reverse well ordered
sums. In fact, as a consequence of Hausdors proof we get that every linear ordrer
is a dense sum of scattered ordered types (see as well [5]).
We shall use the followin notation:
Notation 0.2. The Erd os-Rado arrows.
(1) ()

means that for every set S such that otp(S, <) = and each
coloring c : [S]

, there is an ordinal i < and a subset T S so that

otp(T, <) = and c [T]

= i
(2) ()

means that the statment ()

does not hold

It is easy to show that if = 1 (i.e., the colorings are dened on singletons)
and is nite, then ()

holds in the class of scattered order types. Trying

to generalize it, we encounter with two problems. First, innite amount of colors
poses a limitation (in the case of scattered order types), even when using just
0
colors. Second, dealing with -tuples with > 1 becomes much more complicated.
For the rst problem, ()
1

is exemplied by = 1+(

+) +(

+)
2
+. . .
(recall that if = otp(S, <) then

is otp(S, >)). For the second problem,

(

+ )
2
2
, so we fail even when trying to use pairs.
Nevertheless, one can still prove positive results for innitely many colors and -
tuples, even when dealing with scattered order types. Aiming to these results, we
need again a bit of notation:
Notation 0.3. Square brackets.
(1) []

means that for every set S such that otp(S, <) = and each
coloring c : [S]

, there is an ordinal i < and a subset T S so that

otp(T, <) = and i / c [T]

(2) []

,
means that for every set S such that otp(S, <) = and each
coloring c : [S]

, there is a subset X , [X[ = and a subset

T x S : c(x) X such that otp(T, <) =
The former property in the above denition is a property of omitting a color,
the latter property is the main concern of this paper. Notice that if []

,
and
, then []

,
. Consequently, we may succeed even with innite number
of colors and colorings of -tuples, if we decrease . In particular, []

,1
is
equivalent to ()

.
In the general case (with no restriction to scattered order types) we can get both
positive and negative results. For example, []

,2
was proved by Shelah in [6],
for every innite and any natural number . On the other hand, it is consistent
to have an order type of cardinality
1
, such that []
2
1
as shown by Hajnal
and Komj ath in [2].
Under these considerations, we seek for ZFC theorems in the class of scattered
order types. It was proved in [4] that []
1
,0
for such types. We generalize
9
0
9


r
e
v
i
s
i
o
n
:
2
0
0
9
-
1
0
-
2
5







m
o
d
i
f
i
e
d
:
2
0
0
9
-
1
0
-
2
5


UNIFORMING n-PLACE FUNCTIONS ON WELL FOUNDED TREES 3
it, to yield the relation []

,0
for every . Notice that ()
1
0
, so the
subscript ,
0
is well motivated.
1. Some Denitions and Notation
This paper is a natural continuation of [4] in which Shelah and Komj ath prove
that for any scattered order type and cardinal there exists a scattered order
type such that []
1
,0
. This was proved by a theorem on colourings of well
founded trees. By Hausdors characterization (see [3] and [5] and the introduction
above ) every scattered order type can be embedded in a well founded tree, so we
can deduce a natural generalization of their theorem to the n-ary case, i.e for every
scattered order type , n < , and cardinal there is a scattered order type
such that []
n
,0
.
We start with a few denitions.
Definition 1.1. For an ordinal we dene ds() = : a decreasing se-
quence of ordinals < . By ds() we mean the class of decreasing sequences of
ordinals.
We say T ds() is a tree when T is non-empty and closed under initial
segments. T, S will denote trees. For S T ds() we say that S is a subtree of
T if it is also a tree. We use the following notation:
Notation 1.2. (1) For , ds() by we mean where is
maximal such that = .
(2) For ds() and a tree T ds() we dene

T = : ( T)( =

)
Note that for ds()) and ) T ds() if (lg()1) > sup(0) :
T then

T ds().
Definition 1.3. We dene the following four binary relations on ds():
(1) Let <
1
x
be the two place relation on ds() dened by <
1
x
i one of
the following: ()(() < () and = ) or .
(2) Let <
2
x
be the two place relation on ds() dened by <
2
x
i one of
the following: ()(() < () and = ) or .
(3) <

x
=<
1
x
<
2
x
.
(4) Let <
3
be the two place relation on ds() dened by <
3
i one of
the following holds: or for the maximal such that = if is
even then () < () and if is odd then () > ().
It is easily veried that <
1
x
, <
2
x
and <
3
are complete orders of ds(), and
therefore <

x
is a partial order. The following remark refers to their order types
dened by <
1
x
, <
2
x
and <
3
on ds() or ds().
Observation 1.4. (1) <
1
x
, <
2
x
are well orderings for ds().
(2) (ds(), <
3
) is a scattered linear order type for every ordinal .
(3) Every scattered linear order type can be embedded in (ds(), <
3
) for some
ordinal .
Proof. (1) Let , = A ds(), we dene by induction on n < an
element a
n
in the following manner a
0
= min(0) : A, assume
a
0
, , a
n1
have been chosen so that a
k
: k < n) ds() and for every
9
0
9


r
e
v
i
s
i
o
n
:
2
0
0
9
-
1
0
-
2
5







m
o
d
i
f
i
e
d
:
2
0
0
9
-
1
0
-
2
5


4 ESTHER GRUENHUT AND SAHARON SHELAH
A a
k
: k < n)
2
x
n (if lg() n then n = ). Now choose
a
n
= min(n) : A n = a
k
: k < n), if that set isnt empty.
As the sequence derived in the above manner is a decreasing sequence of
ordinals it is nite, say a
0
, a
n1
have been dened and a
n
cannot be
dened, we will show that a = a
k
: k < n) is the minimal element of A
with respect to <
2
x
. By the denition of the sequence there is an A
so that n = a, if lg() > n then we could have dened a
n
, so = a
and in particular a A, and for every A a we have a <
2
x
. Let
n

= minm : am A so an

is the <
1
x
- minimal element in A.
(2) The proof is by induction on . Assume that (ds(), <
3
) is a scattered
linear order type for every < , and assume towards contradiction
that Q can be embedded in (ds(), <
3
), q
q
. Let C = : (p, q
Q)(
p
() ,=
q
()), = min C and = : (q Q)(
q
() = ).
Without loss of generality is even and for
0
= min ,
1
= min
0

there are q
0
< q
1
Q so that
qi
() =
i
, i = 0, 1. Now (q
0
, q
1
) = B
0
B
1
where B
i
= p (q
0
, q
1
) :
p
() =
i
. For some i 0, 1 the set
B
i
contains an interval of Q and is embedded in (
qi
( + 1)

ds(
i
), <
3
)
but this would imply that Q can be embedded in (ds(
i
), <
3
) which is a
contradiction to the induction hypothesis.
(3) By Hausdors characterization it is enough to show for ordinals and
that both A
,
= (ds(), <
3
) and A
,
= (ds(), <
3
)

can be
embedded in (ds( + 2 + 1), <
3
). The embedding is given as follows,
for (, ) A
,
we have (, ) + + + 1, + )

, and for
(, ) A
,
we have (, ) + 2, + + )

Definition 1.5. For trees T

1
, T
2
ds(), f : T
1
T
2
is an embedding of T
1
into T
2
if f preserves level, and <
1
x
(or equivalently, <
2
x
, <

x
or <
3
).
Observation 1.6. For trees T
1
, T
2
ds(), if f : T
1
T
2
preserves level
and then in order to determine whether f is an embedding it is enough to check
for T
1
and ordinals
1
<
2
such that
i
=

i
) T
1
(i = 1, 2) that
f(
1
) <

x
f(
2
).
As T ds() is well founded, i.e there are no innite branches, it is natural
to dene a rank function. in the following denition rk
T,
isnt the standard rank
function but for = 1 we get a similar denition to the usual denition of a rank
on a well founded tree.
Definition 1.7. For a tree T ds() and cardinal dene rk
T,
() :
ds() 1 Ord by induction on as follows:
(a) rk
T,
() 0 i T.
(b) rk
T,
() + 1 i [ :

) T rk
T,
(

)) [.
(c) rk
T,
() limit i ( < )(rk
T,
() ).
We say that rk
T,
() = i rk
T,
() but rk
T,
() + 1.
Denote rk
T,
(T) = rk
T,
()), and rk
T
() = rk
T,1
().
Definition 1.8. For a tree T ds(), T and cardinals , we dene the
reduced rank rk

T,
() = min, rk
T,
().
We rst note a few properties of the rank function.
9
0
9


r
e
v
i
s
i
o
n
:
2
0
0
9
-
1
0
-
2
5







m
o
d
i
f
i
e
d
:
2
0
0
9
-
1
0
-
2
5


UNIFORMING n-PLACE FUNCTIONS ON WELL FOUNDED TREES 5
Observation 1.9. For T ds() and an ordinal we have:
(1) For cardinals

we have rk
T,
() rk
T,
(), and in particular
rk
T
() rk
T,
()
(2) rk
T
() = rk
T
(

)) + 1 :

) T.
(3) rk
ds()
()) = .
(4) If rk
T,
() , then we can embed

ds() into T, so that

for .
Proof. 3 The proof is by induction on .
For = 0 this is obvious. Assume correctness for every < . ds() =

<
)

: ds(). For every < , ds() we have rk

ds()
()

) =
rk
ds()
(), therefore (the last equality is due to the induction hypothesis):
rk
ds()
()

) + 1 : ds() = rk
ds()
() + 1 : ds()
= rk(ds())
=
We therefore have rk(ds()) = + 1 : < =
4 The proof is by induction on .
For = 0 there is nothing to prove.
Assume correctness for every < , and rk
T,
() , . For
< let C

= : rk
T,
(

)) , so [C

[ and C

for

< < . By induction on < we can choose an increasing sequence

of ordinals

such that

= min

where

= C

: (

<
)( >

). Assume towards contradiction that

is empty, and let

C

< ) C

. For every C

(and there is such as

[C

[ whereas [C

[ [[ < ) as /

then there is

< such that

<

, assume

is minimal with this property, but that contradicts the

choice of

.
By the induction hypothesis for every < there is

which embeds
(

))

ds() in T so that

) = Id. We now dene

ds() T in the following manner, if then

() = , else
=

for some ds(), so there is < such that = )

1
with

1
ds(), and we dene

() =

1
).

obviously preserves level.

For
1

2
in

ds() if
1
then obviously

(
1
)

(
2
), and
otherwise for some < we have
i
=

i
, i 1, 2,
1

2

ds(), and as

is an embedding we have:

(
1
) =

1
)

2
) =

(
2
).
For

ds(),
1
<
2
ordinals such that for i = 1, 2
i
=

i
)

ds(), necessarily and there are

1

2
< ,
i
ds(
i
) so
that
i
=

i
)

i
. If
1
=
2
= then
1
<

x

2
, and as

is an
embedding,

(
1
) =

1
) <

2
) =

(
2
)
On the other hand, if
1
,=
2
then

(
i
)(lg()) =
i
, and as
1
<
2
,
also in this case

(
1
) <

(
2
).
9
0
9


r
e
v
i
s
i
o
n
:
2
0
0
9
-
1
0
-
2
5







m
o
d
i
f
i
e
d
:
2
0
0
9
-
1
0
-
2
5


6 ESTHER GRUENHUT AND SAHARON SHELAH
By Observation 1.6

is an embedding, and by denition

:
= Id.

The following theorem was proved By Komj ath and Shelah in [4]:
Theorem 1.10. Assume is an ordinal and a cardinal. Set = ([[

0
)
+
,
and let F : ds(
+
) . Then there is an embedding : ds() ds(
+
) and a
function c : such that for every ds() of length n + 1
F(()) = c(n).
In what follows we will generalize the above theorem, in the process we will use
innitary logics. For the readers convenience we include the following denitions.
Definition 1.11. (1) For innite cardinals , , and a vocabulary con-
sisting of a list of relation and function symbols and their arity which is
nite, the innitary language L
,
for is dened in a similar manner to
rst order logic. The rst subscript, , indicates that formulas have <
free variables and that we can join together < formulas by
_
or
_
, the
second subscript, , indicates that we can put < quantiers together in
a row.
(2) Given a structure B for we say that A is an L
,
-elementary submodel
(or substructure), and denote A
,
B or A
L
,
B, if A is a substruc-
ture of B in the regular manner, and for any L
,
formula with free
variables and a

[A[ we have
B [= ( a) A [= ( a).
The Tarski-Vaught condition for a substructure A of B to be an elemen-
tary submodel is that for any L
,
-formula with parameters a A we
have
B [= x( x a) A [= x( x a).
(3) A set X is transitive if for every x X we have x X.
(4) For every set X there exists a minimal transitive set, which is denoted by
TC(X), such that X TC(X).
(5) For an innite regular cardinal we dene
H() = X : [TC(X)[ < .
Remark 1.12. In this paper the main use of innitary logic will be in the
following manner:
(1) will consist of the two binary relations and <

, so [L

+
,
+()[ = 2

.
(2) If

and A
,
B then also A

,
B.
(3)
,
is a transitive relation.
(4) For an innite cardinal let =
+
, = 2

, so is regular and
<
= .
Recall that for a structure B and X |B| such that [X[ +
B there is an elementary L
,
submodel A of B of cardinality which
includes X.
For further reference on this point see [1].
(5) If A
,
B and x is denable in B over A (i.e with parameters in A) by
an L
,
-formula, then it is also denable in A by the same formula. In
particular if A
,
B and X [A[, [X[ < then X [A[.
9
0
9


r
e
v
i
s
i
o
n
:
2
0
0
9
-
1
0
-
2
5







m
o
d
i
f
i
e
d
:
2
0
0
9
-
1
0
-
2
5


UNIFORMING n-PLACE FUNCTIONS ON WELL FOUNDED TREES 7
Definition 1.13. We say that two nite sequence

: < n),

: < n) are
similar when:
(a) lg(

) = lg(

) for < n.
(b) lg(

m
) = lg(

m
) for , m < n.
(c) (

<
2
x

m
) (

<
2
x

m
) for , m < n (equivalently, we could use <
1
x
).
Observation 1.14. (1) Similarity is an equivalence relation and the num-
ber of equivalence classes of nite sequences is
0
.
(2)
1
, . . . ,
k
,

),
1
, . . . ,
k
,

) are similar if
(a)
1
<
2
x

2
<
2
x
. . . <
2
x

k
(b)
k
<
2
x

(c)
k
<
2
x

(d) lg(

) = lg(

)
(e) lg(

k
) = lg(

k
)
Proof. (1) Similarity is obviously an equivalence relation.
The equivalence class of a nite sequence of ds() is determined by its
length n, the lengths n
i
: i < n) of its elements, the lengths n
i,j
: i, j <
n) of their intersections, and a permutation of n (the order of the elements
according to <
1
x
). Therefore for each n < there are
0
equivalence
classes of sequences of length n, and so the number of equivalence classes
of nite sequences of ds() is
0
.
(2) We need to show that lg(

i
) = lg(

i
) for every 0 < i < k.

k
<
2
x

and
k
<
2
x

. If

k
then we also have lg(

k
) =
lg(

k
) = lg(

) = lg(

) so

k
, and

. In this case obviously

the required sequences are similar, so we can assume that there is such
that
k
=

and

() >
k
(). By the same reasoning as above we
deduce that
k
=

and

() ,=
k
() so necessarily

() >
k
().

The last term we will need before moving on to the main theorem is that of
uniformity.
Definition 1.15. Let T ds() be a tree, c : [T]
<0
C. We identify
u [T]
<0
with the <
2
x
-increasing sequence listing it.
(1) We say T is c-uniform if for any similar u
1
, u
2
in [T]
<0
we have c(u
1
) =
c(u
2
).
(2) We say T is c-end-uniform (or end-uniform for c) when
if
1
<
2
x

2
<
2
x
. . . <
2
x

k
<
2
x

are in T and lg(

) = lg(

), lg(
k

) = lg(
k

) (equivalently
1
. . .
k
,

),
1
. . .
k
,

) are similar-see
1.4(3))
then c(
1
. . .
k
,

)) = c(
1
, . . . ,
k
,

)).
(3) We say T is c-n-end-uniform (or n-end-uniform for c) when for k < ,

i
,

j
,

j
ds() (0 < i k, 0 < j n) such that

1
<
2
x
<
2
<
2
x
. . . <
2
x

k
<
2
x

1
<
2
x
. . . <
2
x

1
<
2
x
<
2
<
2
x
. . . <
2
x

k
<
2
x

1
<
2
x
< . . . <

n
if those two sequences are similar then
c(
1
. . . ,

1
. . .)) = c(
1
. . .

1
. . .)).
9
0
9


r
e
v
i
s
i
o
n
:
2
0
0
9
-
1
0
-
2
5







m
o
d
i
f
i
e
d
:
2
0
0
9
-
1
0
-
2
5


8 ESTHER GRUENHUT AND SAHARON SHELAH
2. Uniforming n-place functions on T ds()
We are now ready for the main theorem of this paper.
Main Claim 2.1. Given a tree S ds() and a cardinal we can nd a tree
T ds() such that
()
1
for every c : [T]
<0
there is T

T isomorphic to S such that cT

is c-end-uniform.
()
2
[T[ <
|S|
+([S[ + ).
Proof. We assume that [S[, are innite cardinals since one of our main goals
is proving a statement of the form x [y]
n
,0
, otherwise the bound on T has to
be slightly adjusted.
For each S let

=
S
() = otp( S : <
2
x
, <
2
x
),

=
5+1
([S[ + ),

=
3
(

)
+
.
Note that

are the maximal ones, and let >>

<>
, and <

be a well
ordering of H() (see 1.11(5)). By denition, for every , S such that <
2
x

we have

<

, and

<

in the following we examine the relation between

and

for ,= .
Observation 2.2. For <
2
x
we have

.
Proof. Since

+ 1 we have:

=
5+1
([S[ + )

5(+1)+1
([S[ + )
=
5
(

)

3
(

)
++
=
+

Let T := ds(
+

), we will show that T is as required. Obviously T meets

requirement ()
2
, and let c : [T]
<0
. Because of the many details in the
following construction we bring it as a separate lemma.
Lemma 2.3. For S we can choose M

, T

and
,n
T for n < with the
following properties:
(1) M

is an L

,
+

-elementary submodel of B = (H(), , <

).
(2) |M

| = 2

.
(3) S, T, c M

.
(4) M

,
,n
M

for <

x
, n < .
(5) Properties of T

:
(a) T

=
,lg()

where T

is isomorphic to ds(2
2

).
(b) If

and are of the same length then they realize the same
L

,
+

-type over M

.
(6) Properties of the
,n
:
(a)
,n
T is of length n.
(b)
,lg()
M

.
(c) lg() = m < n
,n
(m) / M

.
9
0
9


r
e
v
i
s
i
o
n
:
2
0
0
9
-
1
0
-
2
5







m
o
d
i
f
i
e
d
:
2
0
0
9
-
1
0
-
2
5


UNIFORMING n-PLACE FUNCTIONS ON WELL FOUNDED TREES 9
(d)
,n
T

, and for n lg() has at least

immediate successors in
T

.
(7) If =
1

), then
(a) M

, T

,
,n
M
1
for n < .
(b)
1,n
,
,n
realize the same L

,
+

-type over M

,
,n
: n < , <

x
.
(c)
1,n
=
,n
for n lg (
1
).
(d)
,n
<

x

1,n
for n = lg().
(e)
,lg()
=
,lg1

) for some .
(f) If

=
1

) with

< then

,lg(

)
<

x

,lg()
.
Proof. We show a construction for such a choice by induction on <
1
x
, yes,
<
1
x
not <
2
x
.
As the induction is on <
1
x
the base of the induction is the case = ). First
choose M

,
+

B of cardinality 2

, so that S, T, c M

(this can be done,

see Remark 1.12). The number of L

,
+

formulas ( x, a) where a

+

>
M

(sequences of length <

+

in M

) is (2

= 2

hence the number of

L

,
+

-types over M

is at most

= 2
2

, so we color T = ds(
+

) by

colors, c

: T

, so that for T its color, c

(), codes the L

,
+

-type which
realizes in B over M

. As
((
2
(

))

0
)
+
=
3
(

)
+
=

by Theorem 1.10 there is an embedding of ds(

2
(

)) in T, and dene T

to be its
image, so that types of sequences from T

depend only on their length. We choose

representatives
,n
: 0 < n < ) from each level larger than 0 so that for n > 0

,n
and has at least

immediate successors in T

and satises 6(c). The latter

can be done by cardinality considerations, |M

| = 2

, while the cardinality of

levels in T

is
2
(

). We let
,0
= ).
It is easily veried that for = ) all the requirements of the construction are met.
We now show the induction step.
Assume =
1

1
), lg(
1
) = r, and that we have dened for
1
(and below by
<
1
x
) and we dene for .

1
Let A

= M

,
,n
: n < , <

x
.
For any <

x
if =
1

) for some <

1
then from requirement (7)(a) of
the construction for we have M

M
1
, and also for all n <
,n
M
1
, else
<

x

1
therefore from requirement (4) of the construction for
1
we have for all
n <
,n
M
1
, and M

M
1
. So A

M
1
, and [A

[
1
, so A

is denable
by an L

1
,
+

1
-formula with parameters in M
1
, so we have:

2
A

M
1
, [A

1
, therefore A

M
1
.
For every n < let

3

n
(x) =

1
,n
(x) =
_
( the L

,
+

type which
1,n
realizes over A

)
And let

4
T

= T : B [=
lg()
().
9
0
9


r
e
v
i
s
i
o
n
:
2
0
0
9
-
1
0
-
2
5







m
o
d
i
f
i
e
d
:
2
0
0
9
-
1
0
-
2
5


10 ESTHER GRUENHUT AND SAHARON SHELAH
As the cardinality of the L

,
+

-type of any B over A

is at most 2

which is
less than
1
, for every n < we have that
n
is an L

1
,
+

1
-formula and therefore
T

is denable in M

1
by an L

1
,
+

1
-formula, namely
T

_
T
_

n<
(lg() = n
n
())
_
_
So

5
T

M
1
and for every n < we obviously have
1,n
T

.
Recall that for all n <
1,n
T

1
, so for any T

1
of length n, we have that
realizes the same L

1
,
+

1
-type over M
1
as
1,n
so in particular they realize
the same L

,
+

-type over A

, so T

. For m n
1,n
,
1,m
n are of the
same length, so in particular
m
(x)
n
(xn). If T

, lg = m so B [=
m
()
therefore B [=
n
(n) and therefore also n T

. We summarize:

6
T

is a subtree of T and T

1
T

.
The following point is a crucial one, we show that:

7
rk
T,
1
(
1,n
) >
1
for every n such that lg(
1
) n < .
Assume toward contradiction that rk
T,
1
(
1,m
)
1
for some lg(
1
) m < ,
and dene for each n such that m n < :

n
= rk
T,
1
(
1,n
) and

n
= rk

1
T,
1
(
1,n
)
(see Denitions 1.7 and 1.8). We now prove by induction on n m that
n

1
,
i.e
n
=

n
. For n = m this is our assumption, and assume that it is known for n.
The following can be expressed by L

1
,
+

1
-formulas with parameters in M
1
:

1
: x has rk

1
T,
1
(x) =
n

2
: x has at least
1
immediate successors y in T

with rk

1
T,
1
(y)

n+1

We have B [=
1
(
1,n
), and since T

1
T

(see
6
) we also have B [=
2
(
1,n
).
By the induction hypothesis for
1
we have
1,n
,
1,n+1
n T

1
and as they are the
same length realize the same L

1
,
+

1
-type over M
1
, so B [=
1

2
(
1,n+1
n), or
in more detail, we have that rk

1
T,
1
(
1,n+1
n) =
n
, i.e rk
T,
1
(
1,n+1
n) =
n
,
and
1,n+1
n has at least
1
immediate successors in T

with reduced rank

n+1
,
so by the denition of rank (Denition 1.7) we have
n
>

n+1
. By the induction
hypothesis
n

1
, therefore also

n+1
=
n+1
. In particular we can deduce
that
n+1
<
n
, so having carried out the induction we have an innite decreasing
sequence of ordinals which is a contradiction.
Recall that lg(
1
) = r so lg() = r + 1,

8
Dene
,
=
1,
for r.
By 2.2
1

+

, by
7
rk
T,
1
(
1,r
) >
1
therefore rk
T,
1
(
1,r
) >
+

so
by denition there are Suc
T
(
1,r
) T

satisfying rk
T,
1
()
+

, dening

,r+1
to be one such which is minimal with respect to <
1
x
(this is equivalent to
demanding that (r) is minimal) can be done by an L

1
,
+

1
formula. We therefore
conclude:

9
We can choose
,r+1
Suc
T
(
1,r
) T

M
1
such that
(i) rk
T,
1
(
,r+1
)
+

.
(ii)
,r+1
is minimal under (i) in <
1
x
.
9
0
9


r
e
v
i
s
i
o
n
:
2
0
0
9
-
1
0
-
2
5







m
o
d
i
f
i
e
d
:
2
0
0
9
-
1
0
-
2
5


UNIFORMING n-PLACE FUNCTIONS ON WELL FOUNDED TREES 11
As
,lg()
M
1
and
1,lg()
(lg(
1
)) / M
1
, we have:

10

,lg()
<
1
x

1,lg()
, notice that as they are the same length <
1
x
<

x
.
Now for any =
1

) S where <
1
we have that <

x
and therefore

,r+1
A

(see
1
).
,lg()
,
1,lg()
realize the same L

,
+

-type over A

, and
by requirement (7)(d) of the construction for (lg() = lg()) we have
,lg()
<
1
x

1,lg()
so also
,lg()
<
1
x

,lg()
and as above, as they are the same length
<
1
x
<

x
, and we therefore conclude that:

11
If =
1

) S where <
1
then
,lg()
<

x

,lg()
.
Since [S, t, c,

lg
()
A

[ < 2

by Remark 1.12 we can choose M

so that

12
M

,
+

M
1
, and therefore also M

,
+

B, of cardinality 2

and S, t, c,

lg
()
A

.
By the same remark we can conclude that

13
M

M
1
.
Lastly we choose T

and
,m
for m > lg().
We have already commented that rk
T,
1
(
,lg()
) >
+

, so from Observation 1.9

we can embed
,lg()

ds(
+

) into T

so that for
,lg()
, and denote
one such embedding by , without loss of generality M
1
.
The number of L

,
+

-types over M

is at most

= 2
2

. We color ds(
+

) in

colors, the color of ds(

+

) is determined by the L

,
+

-type which (
,lg()

)
realizes over M

, call this coloring c

. As ((
2
(

))

0
)
+
=
3
(

)
+
=

, we can
use 1.10 to get an embedding of ds(
2
(

)) into ds(
+

) so that for ds(

2
(

))
the L

,
+

-type that
,n+1

() realizes over M

depends only on its length. Since

the set X of L

,
+

-types over M

is in M
1
of cardinality at most

<
1
we have
X M
1
, also ds(
+

) M
1
so c

M
1
and therefore without loss of generality
M
1
. We dene

14
T

=
,lg()

_
ds(
2
(

))
_
.
T

M
1
and meets requirement (5) of the construction. We will now choose rep-
resentatives
m
: 0 < m < ) from each level of ds(
2
(

)) so that
,n+1

(
m
)
has at least

immediate successors in T

and
,n+1

(
m
)(lg()) / M
1
, since
the existence of such representatives in B can be expressed by an L

1
,
+

1
-formula
with parameters in M
1
so without loss of generality
m
M
1
and dene

15

,lg()+m
=
,n+1

(
m
).
T

is a subtree of T

therefore T

realizes the same L

,
+

type over A

as

1,lg()
. The
,n
for n > lg() were chosen to satisfy (6)(c)-(d) so in particular
they are in T

, and therefore realize the same L

,
+

-type over A

as
1,n
. By
the induction hypothesis we have already constructed for
1
so for all n we have
lg(
,n
) = lg(
1,n
) = n so also (6)(a) is satised. Requirements (1)-(4) and (6)(b)
of the construction are taken care of by
12
.
7
-
11
,
13
and
15
guarantee
requirement (7).
All that is left in order to complete the proof of the claim is to show that
,lg()
:
S is end-uniform with respect to c.
Let
1
<
2
x

2
<
2
x
. . . <
2
x

k
<
2
x

, be as in 1.15(2); without loss of generality

9
0
9


r
e
v
i
s
i
o
n
:
2
0
0
9
-
1
0
-
2
5







m
o
d
i
f
i
e
d
:
2
0
0
9
-
1
0
-
2
5


12 ESTHER GRUENHUT AND SAHARON SHELAH

<

. Let t = lg(

),

=
+

and A =
,lg
: <

(t + 1).
We rst show that for every i k
i
<

(t + 1) so that
i.lg(i)
A. As

i
<
2
x

and lg(
i

) = lg(
i

) so

i
, therefore there is
i
such that

i
=

i
and
i
(
i
) <

(
i
), but then
i

i
=

i
i.e

i
=

i
so
i
t
(and
i
(
i
) <

(
i
)) and
i
<

(t + 1).
We now prove by induction on [t, lg(

)] that

,lg
and

t,lg
realize the
same L

,
-type over A. For = t this is obvious. Let us assume correctness for
and prove for +1. For every n < by (7)(b) of the construction

,n
,

(+1),n
realize the same L

(+1)
,
+

(+1)
-type over M

,
,n
: <

( + 1) and
in particular over A, for if <

(t + 1) then also <

( + 1). So

,lg
,

(+1),lg
realize the same L

(+1)
,
+

(+1)
-type so also the same L

,
-
type over A, and from the induction hypothesis

t,lg
and

,lg
realize the
same L

,
-type over A. Similarly we show for

, so

,lg
and

,lg
realize
the same L

1
,
+

1
-type over A.
From the above we can deduce that in particular
c(
1,lg(1)
, . . . ,

k
,lg(
k
)
,

,lg(

)
)) = c(
1,lg(1)
, . . . ,

k
,lg(
k
)
,

,lg(

)
)).

Conclusion 2.4. Given a tree S ds() and n() < and we can nd a
tree T ds() such that:
()
1
For every c : [T]
<0
there is S

T isomorphic to S such that S

is
n()-end-uniform for c.
()
2
In particular, for every c : [T]
n()
is S

T isomorphic to S such that

cS

depends only on the equivalence classes of the equivalence relation

dened in 1.13.
()
3
[T[ <
1,n()
([S[, ) (see Denition 2.5 below).
Proof. Let S, be as above. Since for [S[,
0
we have that
1,n()
([S[,
0
) =

1,n()
([S[, ), replacing with
0
gives the same bound, and we can therefore
assume that =
0
.
Let h
n
: n < ) be the equivalence classes of the similarity relationship on nite
sequences of ds() (see 1.14(1)), and let f :

( 1) be one-to-one and
onto.
We construct by induction a sequence T
n
: n < ) so that T
0
= S, and for every
n > 0:
(a) [T
n
[ <
1,n
([S[, )
(b) T
n1
, T
n
, correspond to S, T, in Theorem 2.1.
(c) For every c : [T
n
]
<0
there is S

T
n
isomorphic to S such that S

is n-end-uniform for c.
By Theorem 2.1 we can obviously construct such a sequence satisfying clauses
(a), (b), We will show by induction on n that for this sequence also clause (c) holds.
For n = 1 this is Theorem 2.1. Assume correctness for n and let c : [T
n+1
]
<0
.
By (b) there is T

T
n+1
isomorphic to T
n
so that T

is end-uniform for c. Let

: T
n
T

be an isomorphism and let d : [T

]
<0

( 1) as follows: for
=
1
. . .
k
) where
1
<
2
x

2
<
2
x
. . . <
2
x

k
and m <
d( )(m) =
_
c(

)) if

) h
m
for some
-1 otherwise
9
0
9


r
e
v
i
s
i
o
n
:
2
0
0
9
-
1
0
-
2
5







m
o
d
i
f
i
e
d
:
2
0
0
9
-
1
0
-
2
5


UNIFORMING n-PLACE FUNCTIONS ON WELL FOUNDED TREES 13
d is well dened as T

is end-uniform for c, and by dening (

1
, . . .
k
) = ((
1
), . . . (
k
))
for
1
, . . .
k
T
n
we have f d : [T
n
]
<0
, so by the induction hypothesis
there is T

T
n
isomorphic to S so that T

is n-end-uniform for f d . We
claim that S

= (T

) is isomorphic to S and that S

is n + 1-end-uniform for c.
As T

is isomorphic to S and is an isomorphism S

is obviously isomorphic to S.
Let the following sequences in S

be similar,

1
<
2
x
<
2
<
2
x
. . . <
2
x

k
<
2
x

1
<
2
x
. . . <
2
x

n+1

1
<
2
x
<
2
<
2
x
. . . <
2
x

k
<
2
x

1
<
2
x
< . . . <

n+1
So in T

the following sequences are similar:

1
(
1
. . .

1
. . .

n
) = (
1
(
1
)
1
(

1
) . . .
1
(

n
))

1
(
1
. . .

1
. . .

n
) = (
1
(
1
)
1
(

1
) . . .
1
(

n
))
so f d (
1
(
1
. . .
k
,

1
. . .

n
)) = f d (
1
(
1
. . .
k
,

1
. . .

n
)). Therefore
we have f(d(
1
. . .
k
,

1
. . .

n
)) = f(d(
1
. . .
k
,

1
. . .

n
)), and as f is one-to-one,
d(
1
. . .
k
,

1
. . .

n
) = d(
1
. . .
k
,

1
. . .

n
), and therefore c(
1
. . .
k
,

1
. . .

n+1
) =
c(
1
. . .
k
,

1
. . .

n+1
), and ()
1
-()
3
are easily veried.
Definition 2.5. For cardinals
0
and dene
1,
(, ) by induction on
.
1,0
(, ) =
0
() = ,
1,+1
(, ) =
1,(,)
+(
1,
(, ) + ), and for a
limit ordinal
1,
(, ) =

<

1,
(, ).
We end with a conclusion for scattered order types.
Conclusion 2.6. For a scattered order type , a cardinal and n < , there
is a scattered order type so that []
n
,0
.
Proof. Given a scattered order type , a cardinal and n < by Observation
1.4(3) we can embed in (ds(), <
3
) for some ordinal . By Conclusion 2.4()
2
above there is an ordinal and a tree T ds() so that for every coloring c :
T
n
there is a subtree S T isomorphic to ds() so that cS depends only on
the equivalence class of similarity. Noting the above Observation, as (T, <
3
) is a
scattered order, and as there are only
0
equivalence classes, we are done.
References
1. M. A. Dickman, Larger innitary languages, Model Theoretic Logics (J. Barwise and S. Fefer-
man, eds.), Perspectives in Mathematical Logic, Springer-Verlag, New York Berlin Heidelberg
Tokyo, 1985, pp. 317364.
2. Andras Hajnal and Peter Komj ath, A strongly non-Ramsey order type, Combinatorica 17,
no.3 (1997), 363367.
3. F. Hausdor, Grundz uge einer Theorie der geordneten Mengen, Math. Ann. 65 (1908), no. 4,
435505. MR MR1511478
4. Peter Komj ath and Saharon Shelah, A partition theorem for scattered order types, Combina-
torics Probability and Computing 12 (2003, no.5-6), 621626, Special issue on Ramsey theory.
math.LO/0212022.
5. Joseph G. Rosenstein, Linear orderings, Pure and Applied Mathematics, vol. 98, Aca-
demic Press Inc. [Harcourt Brace Jovanovich Publishers], New York, 1982. MR MR662564
(84m:06001)
6. Saharon Shelah, Consistency of positive partition theorems for graphs and models, Set theory
and its applications (Toronto, ON, 1987), Lecture Notes in Mathematics, vol. 1401, Springer,
Berlin-New York, 1989, ed. Steprans, J. and Watson, S., pp. 167193.
9
0
9


r
e
v
i
s
i
o
n
:
2
0
0
9
-
1
0
-
2
5







m
o
d
i
f
i
e
d
:
2
0
0
9
-
1
0
-
2
5


14 ESTHER GRUENHUT AND SAHARON SHELAH
Institute of Mathematics The Hebrew University of Jerusalem, Jerusalem 91904,
Israel. e-mail: esth@math.huji.ac.il
Institute of Mathematics The Hebrew University of Jerusalem, Jerusalem 91904,
Israel and Department of Mathematics Rutgers University New Brunswick, NJ 08854,
USA e-mail: shelah@math.huji.ac.il