6

8
6


r
e
v
i
s
i
o
n
:
1
9
9
8
-
0
6
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
8
-
0
6
-
2
3


THE YELLOW CAKE
ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
Abstract. In this paper we consider the following property:
(
Da
) For every function f : R R R there are functions g
0
n
, g
1
n
: R R
(for n < ) such that
(x, y R)(f(x, y) =
X
n<
g
0
n
(x)g
1
n
(y)).
We show that, despite some expectation suggested by [Sh 675], (
Da
) does
not imply MA(centered). Next, we introduce cardinal characteristics of the
continuum responsible for the failure of (
Da
).
0. Introduction
In the present paper we will consider the following property:
(
Da
) For every function f : R R R there are functions g
0
n
, g
1
n
: R R
(for n < ) such that
(x, y R)(f(x, y) =

n<
g
0
n
(x)g
1
n
(y)).
Davies [Da74] showed that CH implies (
Da
) and Miller [Mixx, Problem 15.11],
[Mi91] and Ciesielski [Ci97, Problem 7] asked if (
Da
) is equivalent to CH. It was
shown in [Sh 675, 3] that the answer is negative. Namely,
Theorem 0.1. (1) (See [Sh 675, 3.4]) MA(-centered) implies (
Da
).
(2) (See [Sh 675, 3.6]) If P is the forcing notion for adding
2
Cohen reals
then
P
(
Da
).
The proof of [Sh 675, Conclusion 3.4]) strongly used the assumptions causing an
impression that the property (
Da
) might be equivalent to MA(-centered).
The rst section introduces a strong variant of ccc which is useful in preserving
unbounded families. In the second section we show that (
Da
) does not imply
MA(-centered). Finally, the in next section we show the combinatorial heart of
[Sh 675, Proposition 3.6] and we introduce cardinal characteristics of the continuum
closely related to the failure of (
Da
).
Notation Most of our notation is standard and compatible with that of clas-
sical textbooks on Set Theory (like Bartoszy nski Judah [BaJu95]). However in
forcing we keep the convention that a stronger condition is the larger one.
The rst author thanks the Hebrew University of Jerusalem for support during his visit to
Jerusalem in Summer98 when this research was done.
The research of the second author was partially supported by The Israel Science Foundation.
Publication 686.
1
6
8
6


r
e
v
i
s
i
o
n
:
1
9
9
8
-
0
6
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
8
-
0
6
-
2
3


2 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
Notation 0.2. (1) For two sequences , we write whenever is a proper
initial segment of , and when either or = . The length of
a sequence is denoted by g().
(2) The set of rationals is denoted by Q and the set of reals is called R. The
cardinality of R is called c (and it is refered to as the the continuum).
The dominating number (the minimal size of a dominating family in

in the ordering of eventual dominance) is denoted by d and the unbounded

number (the minimal size of an unbounded family in that order) is called
b.
(3) The quantiers (

n) and (

n) are abbreviations for

(m )(n > m) and (m )(n > m),
respectively.
(4) For a forcing notion P,
P
stands for the canonical Pname for the generic
lter in P. With this one exception, all Pnames for objects in the extension
via P will be denoted with a dot above (e.g.

A,

f).
1. Tsweet forcing notion
Denition 1.1. An uncountable family T

is spread if
() for each k

, n

< and a sequence f

,n
: <
1
, n < n

) of pairwise
distinct elements of T there are an increasing sequence
i
: i < )
1
and an integer k > k

such that
(i < )(n < n

)(f
i,n
(k) < f
i+1,n
(k)).
Remark 1.2. (1) Note that if an uncountable family T

has the property
that its every uncountable subfamily is unbounded on every K []

then
T is spread.
(2) If is uncountable and one adds many Cohen reals c

: < )

then c

: < is a spread family.

(3) If there is a spread family then b =
1
(so in particular MA
2
(centered)
fails).
Denition 1.3. Let T

be a spread family. A forcing notion P is Tsweet if
the following condition is satised:
()
F
sweet
for each sequence p

: <
1
) P there are A []

1
, k

< and a
sequence f
,n
: n < n

, A) T such that (, n) ,= (

, n

) f
,n
,=
f

,n
and
() if
i
: i < ) is an increasing sequence of elements of A such that for
some k (k

, )
(i < )(n < n

)(f
i,n
(k) < f
i+1,n
(k))
then there is p P such that p (

i )(p
i

P
).
Proposition 1.4. Assume that T

is a spread family and P is an Tsweet
forcing notion. Then

P
T is a spread family .
Proof. First note that easily Tsweetness implies the ccc.
6
8
6


r
e
v
i
s
i
o
n
:
1
9
9
8
-
0
6
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
8
-
0
6
-
2
3


THE YELLOW CAKE 3
Suppose that k
+
< ,

f
,n
: <
1
, n < n
+
) are Pnames for elements of T,
p P and
p
P
(,

<
1
)(n, n

< n
+
)((, n) ,= (

, n

)

f
,n
,=

f

,n
).
For <
1
choose conditions p

p and functions f
,n
T (for n < n
+
) such
that p

(n < n
+
)(

f
,n
= f
,n
). Passing to a subsequence, we may assume that
(, n) ,= (

, n

) f
,n
,= f

,n
.
Choose k

> k
+
, a set A []

1
and a sequence f
,n
: A, n
+
n < n

)
as guaranteed by ()
F
sweet
of 1.3 for p

: <
1
) (note that here, for notational
convenience, we use the interval [n
+
, n

) instead of n

there). Shrinking the set

A and possibly decreasing n

(and reenumerating f
,n
s) we may assume that all
functions in appearing in f
,n
: A, n < n

) are distinct. By () of 1.1 we nd

k > k

and an increasing sequence

i
: i < ) A such that
(i < )(n < n

)(f
i,n
(k) < f
i+1,n
(k)).
But it follows from () of 1.3 that now we can nd a condition q P such that
q (

i )(p
i

P
). As all conditions p

are stronger than p we may demand

that q p. Now use the choice of the p
i
s and f
i,n
(for n < n
+
) to nish the
proof.
Theorem 1.5. Assume T is a spread family. Let P

,

Q

: < ) be a nite
support iteration of forcing notions such that for each < we have
(1)
P
T is spread , and
(2)
P

is Tsweet .
Then P

is Tsweet (and consequently,

P
T is a spread family ).
Proof. We show this by induction on .
Case 1: = + 1
Let p

: <
1
) P
+1
. Take a condition p

such that
p

<
1
: p

is uncountable
(there is one by the ccc). Next, use the assumption that

Q

is Tsweet and get

P

names

A [
1
]

1
and

k

, n

and

f
,n
:

A, n < n

) T such that the

condition p

forces that they are as guaranteed by ()

F
sweet
of 1.3 for the sequence
p

() : <
1
, p

).
Let A

be the set of all <

1
such that there is a condition stronger than
both p

and p

which forces that p

() is in

A. Clearly [A

[ =
1
. For each
A

choose a condition q

stronger than both p

and p

which forces
that p

()

A and decides the values of

k

, n

and

f
,n
: n < n

). Next we may
choose A

[A

1
, k

, n

and f
,n
: A

, n < n

) T such that (for each

A

and n < n

) q

= k

& n

= n

&

f
,n
= f
,n
. Moreover we may
demand that the f
,n
s are pairwise distinct (for A

, n < n

).
Apply the inductive hypothesis to the sequence q

: A

) (and P

) to get
A [A

1
, k
+
, n
+
> n

and f
,n
: A, n

n < n
+
). For simplicity we
may assume that there are no repetitions in the sequence f
,n
: A, n < n

)
(we may shrink A and decrease n

reenumerating f
,n
s suitably). We claim that
this sequence and maxk

, k
+
satisfy the demand in () if 1.3. So suppose that
6
8
6


r
e
v
i
s
i
o
n
:
1
9
9
8
-
0
6
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
8
-
0
6
-
2
3


4 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH

i
: i < ) is an increasing sequence of elements of A such that for some k > k

, k
+
we have
(i < )(n < n
+
)(f
i,n
(k) < f
i+1,n
(k)).
Clearly, by our choices, we nd a condition p
+
P

stronger than p

such that
p
+
(

i )(q
i

P

). Next, in V
P

, we look at the sequence p

i
() : q
i

, i < ). We may nd a P

name p
+
() such that (p
+
forces that)
p
+
()

i )(q
i

P

& p
i
()

).
Look at the condition p
+
p
+
().
Case 2: is a limit ordinal.
If p

: <
1
) P

then, under the assumption of the current case, for some

A [
1
]

1
and < , the sets supp(p

) : A are pairwise disjoint. Apply

the inductive hypothesis to P

and the sequence p

: A).
Conclusion 1.6. Suppose that >
1
is a regular cardinal such that
<
= and
( < )(
0
< ). Then there is a ccc forcing notion P of size such that

P
there is a spread family T

of size & c = & MA(Tsweet) .
Proof. First note that if P is an Tsweet forcing notion, 1

P (for < < ) are

dense subsets of P and p P then, under our assumptions, there is a set P

P of
size less than such that p P

and
if p, q P

are incompatible in P

then they are incompatible in P,

if p
i
: i < ) P

is not a maximal antichain in P then it is not in P

,
for each < the intersection 1

is dense in P

.
(Thus P

< P and so it is Tsweet.)

Now, using standard bookkeeping arguments, build a nite support iteration
P

,

Q

: < ) such that

(1) Q
0
is the forcing notion adding many Cohen real T = f

: < )

(with nite conditions), [so in V

Q0
, the family T is spread]
(2) for each < ,
P1+

Q
1+
is a Tsweet forcing notion of size < ,
(3) if

Q is a P

name for a Tsweet forcing notion of size < then for many
< ,

Q is a P

name and
P

Q =

Q

.
It follows from 1.5 that in V
P
(for 0 < ) the family T is spread, so there are
no problems with carrying out the construction. Easily P

is as required.
Remark 1.7. Note the similarity of MA(Tsweet) to the methods used in [Sh:98,
4].
2. More on Davies Problem
The aim of this section is to show that (
Da
) does not imply MA(centered).
Let
n
: n < ) be an enumeration of
>
such that g(
n
) n. For distinct

0
,
1


let (
0
,
1
) = 1 + maxm :
m

0
&
m

1
. (Note that

0
(
0
,
1
) ,=
1
(
0
,
1
).)
Assume that there exists a spread family of size c and let T =

: < c)

be such a family (later we will choose the one coming from adding many Cohen
reals).
Denition 2.1. Let < c be an ordinal and let f : R.
6
8
6


r
e
v
i
s
i
o
n
:
1
9
9
8
-
0
6
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
8
-
0
6
-
2
3


THE YELLOW CAKE 5
(1) A approximation is a sequence g = g

: < 2,
>
) such that:
(a) g

: Q (for < 2,
>
),
(b) if < then ( < c)(

k )(g

k
() ,= 0 &
k

),
(c) if < ,
>
and neither nor
g()
is an initial segment of

,
then g
0

() = g
1

() = 0.
(2) If
0
<
1
and g
k
= g
,k

: < 2,
>
) (for k = 0, 1) are
k

approximations such that g

,0

g
,1

(for all < 2 and

>
) then we
say that g
1
extends g
0
(in short: g
0
_ g
1
).
(3) We say that a approximation g agrees with the function f if
(, < )

f(, ) =

>

g
0

() g
1

() and the series converges absolutely

.
Proposition 2.2. If g

are

approximations (for <

) such that the sequence

g

: <

) is _increasing and

<

then there is a

approximation
g

such that ( <

)( g

_ g

). Moreover, if f :

R and each g

agrees with f(

) then g

agrees with f.
Thus if we want to show that (
Da
) holds we may take a function f : c c R
(it should be clear that we may look at functions of that type only) and try to build
a _increasing sequence g

: < c) of approximations. If we make sure that g

is
a approximation that agrees with f( ) then the limit g
c
of g

s will give us
witnesses for f. (Note that by the absolute convergence demand in 2.1(3) we do
not have to worry about the order in the series.) At limit stages of the construction
we use 2.2, but problems may occur at some successor stage. Here we need to use
forcing.
Denition 2.3. Assume that < c is an ordinal, and f : ( + 1) ( + 1) R.
Let g = g

: < 2,
>
) be a approximation which agrees with f .
We dene a forcing notion P
g,
f
as follows:
a condition is a tuple p = Z
p
, j
p
, r
p
,
: < 2,
j
p
>
)) such that
() j
p
< and Z
p
is a nite subset of , r
p
,
Q (for < 2,
j
p
>
),
() the set
j
p
>
: r
p
0,
,= 0 or r
p
1,
,= 0 is nite, and if
j
p
>
and
neither nor
g()
is an initial segment of

then r
p
,
= 0,
() if Z
p
then
[f(, )

g
0

() r
p
1,
:
j
p
>
[ < 2
j
p
,
[f(, )

r
p
0,
g
1

() :
j
p
>
[ < 2
j
p
, and
[f(, )

r
p
0,
r
p
1,
:
j
p
>
[ < 2
j
p
(note that by demand () all the sums above are nite),
() if , Z
p
are distinct then (

) < j
p
;
the order is dened by p q if and only if
(a) j
p
j
q
, Z
p
Z
q
and r
p
,
= r
q
,
for
j
p
>
, < 2,
6
8
6


r
e
v
i
s
i
o
n
:
1
9
9
8
-
0
6
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
8
-
0
6
-
2
3


6 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
(b) if Z
p
then

[r
p
0,
g
1

()[ :
j
q
>

j
p
>
< 4
1 2
j
p
j
q
2
j
p
1
,

[g
0

() r
p
1,
[ :
j
q
>

j
p
>
< 4
1 2
j
p
j
q
2
j
p
1
, and

[r
p
0,
r
p
1,
[ :
j
q
>

j
p
>
< 4
1 2
j
p
j
q
2
j
p
1
.
Proposition 2.4. Suppose that < c, f : ( + 1) ( + 1) R and g is a
approximation that agrees with f . Then:
(1) P
g,
f
is a (non-trivial) Tsweet forcing notion of size [[ +
0
.
(2) In V
P
g,
f
, there is a ( +1)approximation g

such that g g

and g

agrees
with f.
Proof. (1) First note that (P
g,
f
, ) is a partial order and easily P
g,
f
,= (re-
member that Z
p
may be empty). Before we continue let us show the following claim
that will be used later too.
Claim 2.4.1. For each j < , < and

the sets
1
j
def
= p P
g,
f
: j
p
j,
1

def
= p P
g,
f
: Z
p
, and
1
j

def
= p P
g,
f
: j < j
p
& ( < 2)(k (j, j
p
))(r
p
,

j
,= 0 &
j
)
are dense subsets of P
g,
f
.
Proof of the claim. Let j < , < ,

and p P
g,
f
.
If j j
p
then p 1
j
, so suppose that j
p
< j. Let
m
: m < m

) enumerate Z
p
.
Choose pairwise distinct j
,m
: < 2, m < m

) (j, ) such that

j
,m

and
g

m
j
,m
(
m
) ,= 0 (remember 2.1(1b)). Fix j

> j such that

j
is not an initial
segment of any
m
(for m < m

). Let j
q
= j + maxj
,m
: < 2, m < m

+ j

,
Z
q
= Z
p
and dene r
q
0,
, r
q
1,
as follows.
(1) If
j
p
>
then r
q
,
= r
p
,
.
(2) If
j
q
>

j
p
>

m
j
,m
: m < m

, < 2 then r
q
1,
= 0.
(3) If =

then r
q
0,
, r
q
1,
Q 0 are such that [r
q
0,
r
q
1,
[ < 2
j
p
and
[f(, )

r
p
0,
r
p
1,
:
j
p
>
r
q
0,
r
q
1,
[ < 2
2j
q
.
(4) If =
m
j
0,m
, m < m

then r
q
1,
Q is such that [g
0

(
m
) r
q
1,
[ < 2
j
p
and
[f(
m
, )

g
0

(
m
) r
p
1,
:
j
p
>
g
0

(
m
) r
q
1,
[ < 2
2j
q
;
if =
m
j
1,m
, m < m

then r
q
0,
Q is such that [r
q
0,
g
1

(
m
)[ < 2
j
p
and
[f(,
m
)

r
p
0,
g
1

(
m
) :
j
p
>
r
q
0,
g
1

(
m
)[ < 2
2j
q
.
One easily checks that q = Z
q
, j
q
, r
q
,
: < 2,
j
q
>
)) is a condition in P
g,
f
stronger than p (and q 1
j
).
6
8
6


r
e
v
i
s
i
o
n
:
1
9
9
8
-
0
6
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
8
-
0
6
-
2
3


THE YELLOW CAKE 7
Now suppose that / Z
p
. Take j
0
> j
p
such that ( Z
p
)((, ) < j
0
).
Let
m
: m < m

) enumerate Z
p
and let j
,m
: < 2, m < m

) (j
0
, ) be
pairwise distinct and such that
j
,m

& g

mj
,m
(
m
) ,= 0. Let j

> j
p
be such
that
j
is not an initial segment of any
m
. Put Z
q
= Z
p
, j
q
= j
p
+maxj
,m
:
< 2, m < m

+ j

, and dene r
q
,
like before, with one modication. If
m
=
and =

j
0,m
then r
q
1,
Q is such that [f(, ) g
0

() r
q
1,
[ < 2
2j
q
; if
m
=
and =

j
1,m
then r
q
0,
Q is such that [f(, ) r
q
0,
g
1

()[ < 2
2j
q
.
Similarly one builds a condition q 1
j

stronger than p (just choose j

suitably).

Now we are going to show that P

g,
f
is Tsweet. So suppose that p

: < )
P
g,
f
. Choose A [
1
]

1
such that
Z
p
: A) forms a system with kernel Z,
for each , A, [Z
p
[ = [Z
p

[, j
p
= j
p

and
r
p
,
: < 2,
j
p
>
) = r
p

,
: < 2,
j
p

>
)
(remember 2.3()),
if , A and : Z
p
Z
p

is the order preserving bijection then Z

is the identity on Z and ( Z
p
)(

j
p
=
()
j
p

).
Let k

= j
p
, n

= [Z
p
Z[ for some (equivalently: all) A. For A let
f
,n
: n < n

) enumerate

: Z
p
Z. Clearly there are no repetitions in
f
,n
: n < n

, A). We claim that this sequence is as required in () of 1.3. So

suppose that
i
: i < ) A is an increasing sequence such that for some k > k

we have
(i < )(n < n

)(f
i,n
(k) < f
i+1,n
(k)).
Passing to a subsequence we may additionally demand that for each m < k, for every
n < n

, the sequence f
i,n
(m) : i < ) is either constant or strictly increasing.
For n < n

let k
n
j
p
be such that the sequence f
i,n
k
n
: i < ) is constant
but the sequence f
i,n
(k
n
) : i < ) is strictly increasing. Take j > k such that
if
m
f
i,n
k
n
, n < n

then m < j. Fix an enumeration

m
: m < m

) of
Z
p
0
(so m

= [Z[ + n

) and choose j

, j
,m
> j + 2 with the properties as in the
rst part of the proof of 2.4.1 (with p
0
in the place of p there). Put Z
q
= Z
p
0
and dene j
q
, r
q
,
exactly as there (so, in particular, for each
j>

j
p
0
>

we have r
q
,
= 0). We claim that q (

i )(p
i

P
g,
f
). So suppose
that q

q, i
0
< . Choose i > i
0
such that for each n < n

and k

> k
n
, if

m
= f
i,n
k

then m > j
q

. Moreover, we demand that if k

n
< k

< j
q

, n < n

then r
q

0,f
i
,nk
= r
q

1,f
i
,nk
= 0 (remember 2.3()). Then we have the eect that
(
j
q

>

j
p
i
>
)( < 2)( Z
p
i
Z)(r
q

,
g
1

() = g
0

() r
q

1,
= 0)).
So we may proceed as in the proof of 2.4.1 and build a condition q
+
stronger than
both q

and p
i
.
(2) Let G P
g,
f
be generic over V. For
>
dene
g
,

() = r
p
,
where p G 1
g()+1
,
g
,

() = g

() for < .
6
8
6


r
e
v
i
s
i
o
n
:
1
9
9
8
-
0
6
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
8
-
0
6
-
2
3


8 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
It follows immediately from 2.4.1 (and the denition of the order on P
g,
f
) that the
above conditions dene a + 1approximation g

= g
,

: < 2,
>
) which
agrees with f and extends g.
Theorem 2.5. Assume that is an uncountable cardinal such that
<
= . Then
there is a ccc forcing notion P of size such that

P
(
Da
) + c = + there is a spread family of size c .
Proof. Using standard bookkeeping argument build inductively a nite support
iteration P

,

Q

: < ) and sequences

: < ),

: < ) and

f

: < )
such that:
(1) Q
0
is the forcing notion adding many Cohen reals

: < )
>
(by
nite approximations; so, in V
Q0
, c = and the family T =

: < is
spread; we use it in the clauses below),
(2)

< ,

f

is a P

name for a function from (

+1) (

+1) to R,

g

is
a P

name for a

approximation (for the family T added by Q

0
) which
agrees with

f

),
(3)
P1+

Q
1+
= P
g,

f
(for T),
(4) if

f is a P

name for a function from ( + 1) ( + 1) to R, < and

g
is a P

name for a approximation which agrees with

f( ) then for
some < , > we have

g =

g

,

f =

f

, =

.
Clearly P

is a ccc forcing notion (with a dense subset) of size . It follows from

2.4(2), 2.2 that
P
(
Da
) (and clearly
P
c = ). Moreover, by 2.4(1), 1.5 we
know that, in V
Q0
, for each [1, ] the forcing notion P

[1, ) is Tsweet, so

P
T is a spread family of size
(by 1.4).
3. When (
Da
) fails.
In this section we will strengthen the result of [Sh 675, 3.6] mentioned in 0.1(2)
giving its combinatorial heart.
Denition 3.1. (1) For a function h such that dom(h) A } and rng(h)
Z and a positive integer n we dene
(h, n) = min[/
0
[ +[/
1
[ : /
0
T(A) & /
1
T(}) &
(w [A]
n
)(A /
0
)(w A) &
(w [}]
n
)(A /
1
)(w A) &
(A
0
/
0
)(A
1
/
1
)(h[A
0
A
1
] ,= Z).
If A = } and h is as above, and n is a positive integer then we dene

(h, n) = min[/[ : / T(A) & (w [A]

n
)(A /)(w A) &
(A /)(h[A A] ,= Z) .
(2) For c = c
n
: n < )

R and

d = d
n
: n < )

R let h

( c,

d) =

n<
c
n
d
n
(dened if the series converges).
We will deal with the following variant of the property (
Da
).
6
8
6


r
e
v
i
s
i
o
n
:
1
9
9
8
-
0
6
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
8
-
0
6
-
2
3


THE YELLOW CAKE 9
Denition 3.2. For a function h :

R

R R let (
Da
h
) mean:
(
Da
h
) For each f : R R R there are functions g
0
n
, g
1
n
: R R (for n < )
such that
(x, y R)

f(x, y) = h(g
0
n
(x) : n < ), g
1
n
(y) : n < ))

.
(So (
Da
) is (
Da
h

), where h

is as dened in 3.1(2).)
Proposition 3.3. Assume that a function h :

R

R R is such that on of
the following condition holds:
(A) (h, 1) < 2
(h,1)
= c, or
(B) (h, 1) < c for some regular cardinal , or
(C)

(h, 2) < c for some regular cardinal .

Then (
Da
h
) fails.
Proof. First let us consider the case of the assumption (A). Let /
0
, /
1
T(

R)
exemplify the minimum in the denition of (h, 1), /

= A

: < (h, 1) (we

allow repetitions). Choose a sequence r

: < (h, 1)) of pairwise distinct reals

and x enumerations s

: < c) of R and

: < c) of
(h, 1)
(h, 1). Let
f : R R R be such that
( < c)( < (h, 1))

f(s

, r

) / h[A
0

A
1
()
]

.
We claim that the function f witnesses the failure of (
Da
h
). So suppose that
g
0
n
, g
1
n
: R R. For < (h, 1) let

b

= g
1
n
: n < )

R and let () < (h, 1)
be such

b

A
1
()
. Take < c such that =

and let a

= g
0
n
() : n < ). Fix

< (h, 1) such that a

A
0

and note that h( a

) h[A
0

A
1
(0)
], so
f(s

, r

) ,= h( a

) = h(g
0
n
(s

) : n < ), g
1
n
(r

) : n < )).
Suppose now that we are in the situation (B). Let c
0
, c
1
:
+

+
(h, 1) be
such that for any sets X
0
, X
1
[
+
]

+
we have
(
0
,
1
< (h, 1))(
0
,
1
) X
0
X
1
)(c
0
(
0
,
1
) =
0
& c
1
(
0
,
1
) =
1
)
(see e.g. [Sh:g, ch III]). Let /
0
, /
1
T(

R) exemplify (h, 1), /

= A

: <
(h, 1) (with possible repetitions). Choose a sequence r

: <
+
) of pairwise
distinct reals and a function f : R R R such that
(
0
,
1
<
+
)

f(r
0
, r
1
) / h[A
0
c0(0,1)
A
1
c1(0,1)
]

.
Now suppose that g
0
n
, g
1
n
: R R and let a

= g

n
(r

) : n < ). Choose
X
0
, X
1
[
+
]

+
and
0
,
1
< (h, 1) such that a

whenever X

. Take

(for < 2) such that c

0
(
0
,
1
) =
0
, c
1
(
0
,
1
) =
1
. Then h( a
0
0
, a
1
1
)
h[A
0
c0(0,1)
A
1
c2(0,1)
], so f(r
0
, r
1
) ,= h(g
0
n
(r
0
) : n < ), g
1
n
(r
1
) : n < )).
Now, suppose that the assumption (C) holds. Let A

: <

(h, 2) be a family
witnessing the minimum in the denition of

(h, 2). Take a function c :

+

(h, 2) such that for every X [

+
]

+
and <

(h, 2) there are

0
<
1
, both in X, such that c(
0
,
1
) = (see e.g. [Sh:g, ch III]). Take a sequence
r

: <
+
) of distinct reals and dene a function f : R R R so that
(
0
,
1
<
+
)(f(r
0
, r
1
) / h[A
c(0,1)
A
c(0,1)
]).
6
8
6


r
e
v
i
s
i
o
n
:
1
9
9
8
-
0
6
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
8
-
0
6
-
2
3


10 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
Like before, suppose that g
0
n
, g
1
n
: R R and let a

= g

n
(r

) : n < ). For each

<
+
there is

(h, 2) such that a

0

, a
1

. Take a set X [
+
]

+
and

<

(h, 2) such that ( X)(

). Then choose
0
<
1
both in X so that
c(
0
,
1
) =

. By our choices, a
0
0
, a
1
1
A
c(0,1)
and h( a
0
0
, a
1
1
) A
c(0,1)
. But
this implies that h(g
0
n
(r
0
) : n < ), g
1
n
(r
1
) : n < )) ,= f(r
0
, r
1
).
Now the phenomenon of [Sh 675, 3.6] is described in a combinatorial way by 3.3,
if one notices the following observation.
Proposition 3.4. Let h :

R

R

R be a function with an absolute denition
(with parameters from the ground model). Suppose that P = P

,

Q

: <
1
) is a
nite support iteration of non-trivial forcing notions. Then for each 0 < n <

P
1
(h, n) =

(h, n) =
1
.
Proof. Work in V
P
2
. For <
1
let A

= V
P

R. Clearly

R =

<1
A

and
for each , <
1
we have h[A

] ,=

R (remember that the function h has
denition with parameters in the ground model; at each limit stage of the iteration
Cohen reals are added).
4. Concluding remarks
One can notice some similarities between the property ()
Da
and the rectangle
problem.
Denition 4.1. (1) Let
2
be the family of all rectangles in RR, i.e. sets of
the form AB for some A, B R. Let B(
2
) be the algebra of subsets of
RR generated by the family
2
and let B

(
2
) be dened inductively by:
B
0
(
2
) consists of all elements of
2
and their complements, B

(
2
) =

<
B

(
2
) for limit , and B
+1
(
2
) is the collection of all countable
unions

n<
A
n
such that each A
n
is in B

(
2
) and of the complements of
such unions. (So B(
2
) = B
1
(
2
).)
(2) Let us introduce the following properties of the family of subsets of RR:
(
Ku
) T(R R) = B(
2
),
(
Ku

) T(R R) = B

(
2
)
Kunen [Ku68, 12] showed the following.
Theorem 4.2. (1) (See [Ku68, Thm 12.5]) MA implies (
Ku
2
).
(2) (See [Ku68, Thm 12.7]) If P is the forcing notion for adding
2
Cohen
reals then
P
(
Ku
).
The relation between (
Da
) and (
Ku
) is still unclear, though the rst implies
the second.
Proposition 4.3. (
Da
) (
Ku

)
Proof. Suppose that A RR and let f : RR 2 be it characteristic function.
Let g
0
n
, g
1
n
be given by (
Da
) for the function f. For a rational number q, n <
and < 2 put
A

q,n
def
= x R : g

n
(x) < q.
6
8
6


r
e
v
i
s
i
o
n
:
1
9
9
8
-
0
6
-
2
3







m
o
d
i
f
i
e
d
:
1
9
9
8
-
0
6
-
2
3


THE YELLOW CAKE 11
It should be clear that the set A can be represented as a Boolean combination
of nite depth of rectangles A
0
q,n
A
1
q

,n
(we do not try to safe on counting the
quantiers).
The following questions arise naturally in this context.
Problem 4.4. (1) Does (
Ku

) (or (
Ku
)) imply (
Da
)?
(2) Is it consistent that for some countable limit ordinal we have (
Ku
+1
) but
(
Ku

) fails?
References
[BaJu95] Tomek Bartoszy nski and Haim Judah. Set Theory: On the Structure of the Real Line.
A K Peters, Wellesley, Massachusetts, 1995.
[Ci97] Krzysztof Ciesielski. Set theoretic real analysis. Journal of Applied Analysis, 3:143190,
1997.
[Da74] R. O. Davies. Representation of functions of two variables as sums of rectangular func-
tions I. Fundamenta Mathematicae, pages 177183, 1974.
[Ku68] Kenneth Kunen. Inaccessibility properties of cardinals. PhD thesis, Stanford University,
1968.
[Mi91] Arnold W. Miller. Arnie Millers problem list. In Haim Judah, editor, Set Theory of
the Reals, volume 6 of Israel Mathematical Conference Proceedings, pages 645654.
Proceedings of the Winter Institute held at BarIlan University, Ramat Gan, January
1991.
[Mixx] Arnold W. Miller. Some interesting problems. circulated notes; available at
http://www.math.wisc.edu/miller.
[Sh:98] Saharon Shelah. Whitehead groups may not be free, even assuming CH. II. Israel Journal
of Mathematics, 35:257285, 1980.
[Sh:g] Saharon Shelah. Cardinal Arithmetic, volume 29 of Oxford Logic Guides. Oxford Uni-
versity Press, 1994.
[Sh 675] Saharon Shelah. On Ciesielskis Problems. Journal of Applied Analysis, 3(2):191209,
1997. math.LO/9801155.
Department of Mathematics and Computer Science, Boise State University, Boise
ID 83725, USA, and Mathematical Institute of Wroclaw University, 50384 Wroclaw,
Poland
E-mail address: roslanow@math.idbsu.edu
URL: http://math.idbsu.edu/roslanow
Institute of Mathematics, The Hebrew University of Jerusalem, 91904 Jerusalem,
Israel, and Department of Mathematics, Rutgers University, New Brunswick, NJ 08854,
USA
E-mail address: shelah@math.huji.ac.il
URL: http://www.math.rutgers.edu/shelah