5

1
6


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







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


Coloring nite subsets of uncountable sets
P. Komj ath

E otv os University
S. Shelah

Hebrew University
October 6, 2003
Abstract
It is consistent for every 1 n < that 2

= n and there is a
function F : [n]
<
such that every nite set can be written at most
2
n
1 ways as the union of two distinct monocolored sets. If GCH holds,
for every such coloring there is a nite set that can be written at least
1
2

n
i=1
_
n+i
n
__
n
i
_
ways as the union of two sets with the same color.
0 Introduction
In [6] we proved that for every coloring F : [
n
]
<
there exists a set
A [
n
]
<
which can be written at least 2
n
1 ways as A = H
0
H
1
for some
H
0
= H
1
, F(H
0
) = F(H
1
) and that for n = 1 there is in fact a function F for
which this is sharp. Here we show that for every n < it is consistent that
2

=
n
and for some function F as above for every nite set A there are at
most 2
n
1 solutions of the above equation. We use historic forcing which was
rst used in [1] and [7] then in [5] and [4]. Under GCH, we improve the positive
result of [6] by showing that for every F as above some nite set can be written
at least T
n
=
1
2

n
i=1
_
n+i
n
__
n
i
_
ways as the union of two sets with the same F
value.
With the methods of [6] it is easy to show the following corollary of our
independence result. It is consistent that 2

=
n
and there is a function
f : R such that if x is a real number then x cannot be written more than
2
n
1 ways as the arithmetic mean of some y = z with f(y) = f(z). ((y, z) and
(z, y) are not regarded distinct.) Another idea of [6] can be used to modify our
second result to the following. If GCH holds and V is a vector space over the
rationals with |V | =
n
, f : V then some vector can be written at least T
n
ways as the arithmetic mean of two vectors with the same f-value.

Supported by the Hungarian OTKA Grant No.T014105

No. 516. Research supported by the Basic Research Foundation of the Israel Academy of
Sciences and Humanities
1
5
1
6


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







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


Notation We use the standard set theory notation. If S is a set, a cardinal,
then [S]

= {A S : |A| = }, [S]
<
= {A S : |A| < }, [S]

= {A
S : |A| }. P(S) is the power set of S. If f is a function, A a set, then
f[A] = {f(x) : x A}.
1 The independence result
Theorem 1 For 1 n < it is consistent that 2

=
n
and there is a function
F : [
n
]
<
such that for every A [
n
]
<
there are at most 2
n
1 solutions
of A = H
0
H
1
with H
0
= H
1
, F(H
0
) = F(H
1
).
For <
n
x a bijection

: ||. For x [
n
]
<
dene
i
(x) for
i < k = min(n, |x|) as follows.
0
(x) = max(x).

i+1
(x) =
1
0(x)
_

i
_

0(x)
[x
0
(x)]
_
_
.
(x) = {
0
(x), . . . ,
k1
(x)}.
So, for example, if n = 0 then (x) = , if n = 1, x = , then (x) =
{
0
(x)} = {max(x)}.
Lemma 1 Given s [
n
]
n
there are at most countably many x [
n
]
<
such
that (x) = s.
Proof By induction on n.
Let (s) =

{x : (x) s}, a countable set for s [

n
]
<
.
Denition The two sets x, y [
n
]
<
are isomorphic if the structures (x; <
,
0
(x), . . . ,
k1
(x)), (y; <,
0
(y), . . . ,
k1
(y)), are isomorphic, i.e., |x| = |y|
and the positions of the elements
i
(x),
i
(y) are the same.
Notice that for every nite j there are just nitely many isomorphism types
of j-element sets.
The elements of P, the applied notion of forcing will be some structures of
the form p = (s, f) where s [
n
]
<
and f : P(s) .
The only element of P
0
is 1
P
= (, , 0), it will be the largest element of
P. The elements of P
1
are of the form p = ({}, f) where f() = 0 = f({}) for
<
n
.
Given P
t
, p = (s, f) is in P
t+1
if the following is true. s = a b is a
disjoint decomposition. p

= ( a, f

) and p

= ( b, f

) are in P
t
where
f

= f|P(a), f

= f|P(b). There is : a b, an isomorphism

2
5
1
6


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







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


between (a, <, P(a), f

) and (b, <, P(b), f

). | is the identity.
For H a the sets H and [H] are isomorphic. a () = b () = .
f f

is one-to-one and takes only values outside Ran(f

) (which is the
same as Ran(f

)). P =

{P
t
: t < }. We make p p

, p

and the ordering

on P is the one generated by this.
Lemma 2 (P, ) is ccc.
Proof Assume that p

P ( <
1
). We can assume by thinning and using the
-system lemma and the pigeon hole principle that the following hold. p

P
t
for the same t < . p

= ( a

, <, P( a

), f

) where the structures

( a

, <, f

) and ( a

, <, f

) are isomorphic for , <

1
, {, a

:
<
1
} pairwise disjoint. We can also assume that if is the isomorphism
between ( a

, <, f

) and ( a

, <, f

) then H and [H] are isomorphic

for H a

. Moreover, if we assume that occupies the same positions

in the ordered sets a

( <
1
) then will be the identity on . As
() is countable, by removing countably many indices we can also assume
that () a

= for <
1
. Now any p

and p

are compatible as we can

take p = ( a

, <, P( a

), f) p

, p

where f f

, f

is an
appropriate extension, i.e., f f

is one-to-one and takes values outside

Ran(f

).
Lemma 3 If (s, f) P, H
0
, H
1
s have f(H
0
) = f(H
1
) then H
0
, H
1
are
isomorphic.
Proof Set (s, f) P
t
. We prove the statement by induction on t. There is
nothing to prove for t < 2. Assume now that (s, f) P
t+1
, s = a b,
: a b as in the denition of (P, ). As f(H
0
) is a value taken twice
by f, both H
0
and H
1
must be subsets of either a or b. We are done by
induction unless H
0
a and H
1
b (or vice versa). Now H
0
and [H
0
]
are isomorphic and f(H
0
) = f([H
0
]) = f(H
1
) so by the inductive hypothesis
[H
0
] and H
1
are ismorphic and then so are H
0
, H
1
.
Lemma 4 If (s, f) P, H
0
, H
1
s, f(H
0
) = f(H
1
), x H
0
H
1
then x
occupies the same position in the ordered sets H
0
, H
1
.
Proof Similarly to the proof of the previous Lemma, by induction on t, for
(s, f) P
t
. With similar steps, we can assume that (s, f) = ( a b, f)
( a, f

), ( b, f

), H
0
a, H
1
b. Notice that x . Now, as
(x) = x, x is a common element of [H
0
] and H
1
and also f

([H
0
]) = f

(H
1
).
3
5
1
6


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







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


By induction we get that x occupies the same position in [H
0
] and H
1
so by
pulling back we get that this is true for H
0
and H
1
.
Lemma 5 If (s, f) P, A s, 0 j n then A can be written at most
2
j
1 ways as A = H
0
H
1
with H
0
, H
1
distinct, f(H
0
) = f(H
1
), and |(H
0
)
(H
1
)| n j.
Proof By induction on j and inside that induction, by induction on t, for
(s, f) P
t
. The case t < 2 will always be trivial.
Assume rst that j = 0. In this case our Lemma reduces to the following
statement. There are no H
0
= H
1
such that (H
0
) = (H
1
). In the inductive
argument we assume as usual that s = ab and so (s, f) P
t+1
was created
from ( a, f

) and ( b, f

), H
0
a, H
1
b. As (H
0
) = (H
1
),
(H
0
) , but then, as () a = , H
0
can have no points outside and
similarly for H
1
, so we can go back, say to (a, f

) P
t
which concludes the
argument.
Assume now that the statement is proved for j and we have p = (s, f) P
t+1
,
s = a b and p was created from p

= ( a, f

) and p

= ( b, f

).
In A a b we can assume that y = A a = , z = A b = as
otherwise we can pull back to p

or p

. But then, if A = H
0
H
1
, then,if,
say, H
0
a, H
1
b hold, then necessarily H
0
a = y, H
1
b = z,
so H
0
= x
0
y, H
1
= x
1
z where x
0
x
1
= x = A . We can create
decompositions of B = x[y] z by taking B = [H
0
] H
1
. But some of these
decompositions will not be dierent and it may happen that we get non-proper
(i.e., one-piece) decomposition. This can only happen if [y] = z, and then
the two decompositions A = (x
0
y) (x
1
z) and A = (x
1
y) (x
0
y)
produce the same decomposition of B, namely, B = (x
0
z) (x
1
z) and there
is but one decomposition, A = (x y) (x z) which cannot be mapped to a
decomposition of B. If this (i.e., [y] = z) does not happen, we are done by
induction. If this does happen, we know that (H
0
) = (x
0
y) has an element
in y (by the argument at the beginning of the proof). As f(x
0
y) = f(x
1
z),
by Lemmas 3 and 4, both H
0
= x
0
y and H
1
= x
1
z have an element
in the -subset, at the same positions which are mapped onto each other by
. We get that (x
0
z) (x
1
z) has at least n j element, so by our
inductive assumption we have at most 2
j
1 decompositions, which gives at
most 2 (2
j
1) + 1 = 2
j+1
1 decompositions of A.
Let G P be a generic subset. Set S =

{s : (s, f) G}, F =

{f :
(s, f) G}.
Lemma 6 There is a p P such that p |S| =
n
.
4
5
1
6


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







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


Proof Otherwise 1 sup(S) <
n
. By ccc, there is an ordinal <
n
for
which 1 sup(S) < , but this is impossible as there are conditions in P
1
forcing that S.
Now we can conclude the proof of the Theorem. If G is generic, and p G
with the condition p of Lemma 6, then in V [G] F witnesses the theorem by
Lemma 5 (for j = n) on the ground set S. As |S| =
n
we can replace it by
n
.
2 The GCH result
Set
T
n
=
1
2
n

i=1
_
n + i
n
__
n
i
_
.
So T
1
= 1, T
2
= 6, T
3
= 31. In general, T
n
is asymptotically c(3 + 2

2)
n
/

n
for some c.
Theorem 2 (GCH) If F : [
n
]
<
then some A [
n
]
<
has at least T
n
decompositions as A = H
0
H
1
, H
0
= H
1
, F(H
0
) = F(H
1
).
Proof By the Erd os-Rado theorem (see [2, 3]) there is a set {x

: <
1
}
which is (n 1)-end-homogeneous, i.e., for some g : [
1
]
<
, if
1
< <

k
<
1
< <
n1
<
1
then
f({x
1
, . . . , x

k
, x
1
, . . . , x
n1
}) = g(
1
, . . . ,
k
).
Select S
1
[
1
]
1
in such a way that g() = c
0
for S
1
. Set
1
= min(S
1
).
In general, if
i
, S
i
are given (1 i < n) pick S
i+1
[S
i
(
i
+ 1)]
1
so that
g(
1
, . . . ,
i
, ) = c
i
for S
i+1
and set
i+1
= min(S
i+1
). Given
1
, . . . ,
n
and S
n
let
n+1
, . . . ,
2n
be the n least elements of S
n
(
n
+ 1).
Our set will be A = {x
1
, . . . , x
2n
}. For 0 i < n the color of any (n + i)-
element subset of A containing x
1
, . . . , x
i
will be c
i
. We can select
1
2
_
2ni
n
__
n
i
_
dierent pairs of those sets which cover A. In toto, we get T
n
decompositions
of A.
References
[1] J. E. Baumgartner, S. Shelah. Remarks on superatomic Boolean alge-
bras, Annal of Pure and Applied Logic 33 (1987), 109129.
[2] P. Erd os, A. Hajnal, A. M ate, R. Rado. Combinatorial Set Theory:
Partition Relations for Cardinals, North-Holland, Studies in Logic,
106, (1984).
5
5
1
6


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







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


[3] P. Erd os, R. Rado. A partition calculus in set theory, Bull. Amer.
Math. Soc. 62 (1956), 427289.
[4] M. Gilchrist, S. Shelah. On identities of colorings of pairs for
n
, sub-
mitted.
[5] P. Komj ath. A set mapping with no innite free subsets, Journal of
Symbolic Logic 56 (1991), 14001402.
[6] P. Komj ath, S. Shelah. On uniformly antisymmetric functions, Real
Analysis Exchange 19 (19931994), 218225.
[7] S. Shelah, L. Stanley: A theorem and some consistency results in
partition calculus, Annal of Pure and Applied Logic 36 (1987), 119
152.
Peter Komj ath
Department of Computer Science
E otv os University
Budapest, M uzeum krt. 68
1088, Hungary
e-mail: kope@cs.elte.hu
Saharon Shelah
Institute of Mathematics
Hebrew University
Givat Ram, 91904
Jerusalem, Israel
e-mail: shelah@math.huji.ac.il
6