Sie sind auf Seite 1von 20

4

3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


Closed measure zero sets
Tomek Bartoszynski

Boise State University


Boise, Idaho
and
Hebrew University
Jerusalem
Saharon Shelah

Hebrew University
Jerusalem
October 6, 2003
Abstract
We study the relationship between the -ideal generated by closed
measure zero sets and the ideals of null and meager sets. We show that
the additivity of the ideal of closed measure zero sets is not bigger than
covering for category. As a consequence we get that the additivity of the
ideal of closed measure zero sets is equal to the additivity of the ideal of
meager sets.
1 Introduction
Let / and ^ denote the ideals of meager and null subsets of 2

respectively
and let c be the -ideal generated by closed measure zero subsets of 2

. It is
clear that c is a proper subideal of / ^.
For an ideal of subsets of 2

dene
1. add() = min[/[ : / &

/ , ,
2. cov() = min[/[ : / &

/ = 2

,
3. unif() = min[X[ : X 2

& X , and
4. cof () = min[/[ : / & B A / B A.
We can further generalize these denitions and put for a pair of ideals 1 ,

The author thanks the Lady Davis Fellowship Trust for full support

Research partially supported by Basic Research Fund, Israel Academy of Sciences, publi-
cation 439
1
4
3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


1. add(1, ) = min[/[ : / 1 &

/ , ,
2. cof (1, ) = min[/[ : / & B 1 A / B A.
Let 1
0
be the ideal of nite subsets of 2

. Note that cov() = cof (1


0
, ),
unif () = add(1
0
, ), add() = add(, ) and cof () = cof (, ).
The goal of this paper is to study the relationship between the cardinals
dened above for the ideals /, ^ and c. We will show that add(/) = add(c)
and cof (/) = cof (c).
It will follow from the inequalities add(c, ^) cov(/) and cof (c, ^)
unif (/) which will be proved in section 3.
Finally in the last section we will present some consistency results we will
show the cov(c) may not be equal to maxcov(^), cov(/) and similarly
unif (c) does not have to be equal to minunif(/), unif (^).
For f, g

let f

g be the ordering of eventual dominance.


Recall that b is the size of the smallest unbounded family in

and d is the
size of the smallest dominating family in

.
Through this paper we use the standard notation.
denotes the standard product measure on 2

. For a tree T 2
<
let [T] be
the set of branches of T. If T is nite (or has terminal nodes) then [T] denotes
the clopen subset of 2

determined by maximal nodes of T. Let m(T) = ([T])


in both cases.
If s T 2
<
then T[s] = t : s

t T where s

t denotes the concate-


nation of s and t. ZFC

always denotes some nite fragmet of ZFC suciently


big for our purpose.
We will conclude this section with several results concerning the cardinal
invariants dened above.
Theorem 1.1 (Miller [Mi])
1. add(/) = mincov(/), b and cof (/) = maxunif (/), d,
2. add(c, /) b and cof (c, /) d. In particular add(c) b and
cof (c) d,
3. cov(/) add(c, ^) and unif (/) cof (c, ^).
We will also use the combinatorial characterizations of cardinals cov(/)
and unif (/).
Theorem 1.2 (Bartoszynski [Ba1])
1. cov(/) is the size of the smallest family F

such that
g

f F

n f(n) ,= g(n).
2. unif(/) is the size of the smallest family F

such that
g

f F

n f(n) = g(n).
2
4
3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


2 Combinatorics
In this section we will prove several combinatorial lemmas which will be needed
later. The following theorem uses the technique from [Ba2].
Theorem 2.1 Suppose that F

: < < add(c, ^) is a family of closed


measure zero sets. Then there exists a partition of into intervals

I
n
: n
and a sequence T
n
: n such that for all n, T
n
2

In
, [T
n
[ 2
|

In|
2
n
and
_
<
F

x 2

n x

I
n
T
n
.
Furthermore, we can require that
<

n F

I
n
T
n
where F

I
n
= s 2

In
: x F

I
n
= s.
Proof Note that if the sequences

I
n
: n and T
n
: n satisfy the
above conditions then the set x 2

n x

I
n
T
n
has measure zero.
For < and n dene
F
n

= x 2

: s 2
n
s

x( n) F

.
By the assumption there exists a measure zero set H 2

such that

<

n
F
n

H.
Lemma 2.2 (Oxtoby [O]) There exists a sequence of nite sets H
n
: n )
such that H
n
2
n
,

n=1
[H
n
[ 2
n
< and H x 2

n xn H
n
.
Proof Since H has measure zero there are open sets G
n
: n ) covering
H such that (G
n
) < 2
n
for n . Represent each set G
n
as a disjoint union
of open basic intervals
G
n
=

_
m=1
[s
n
m
] for n .
Let H
n
= s 2
n
: s = s
k
l
for some k, l for n . It follows that

n=1
[H
n
[ 2
n

n=1
(G
n
) 1. If x H then x

n
G
n
. Therefore
xn F
n
must hold for innitely many n.
Therefore
_
<
_
n
F
n

x 2

n xn H
n
.
For every < dene an increasing sequence k

n
: n ) as follows: k

0
= 0
and for n ,
k

n+1
= min
_
_
_
m : F
k


m
_
j=k

n
[H
j
]
_
_
_
.
3
4
3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


Since sets F
n

are compact this denition is correct.


We will need an increasing sequence k
n
: n ) such that
<

n m k
2n
< k

m
< k

m+1
< k
2n+1
and
2
kn

j=kn+1
[H
j
[
2
j

1
2
n
.
To construct such a sequence we will use the following lemma:
Lemma 2.3 Suppose that M [= ZFC

and [M[ < d. Then there exists a func-


tion g

such that either


f M

n m g(2n) < f(m) < f(m+ 1) < g(2n + 1)


or
f M

n m g(2n + 1) < f(m) < f(m+ 1) < g(2n + 2).


Proof Let g

be an increasing function such that g ,

f for f M

.
We will show that g has required properties.
Suppose not. Let f
1
, f
2
M

be such that for all n,


[[g(2n), g(2n + 1)] ran(f
1
)[ 1 and [[g(2n + 1), g(2n + 2)] ran(f
2
)[ 1.
We will get a contradiction by constructing a function f M

which
dominates g.
Dene f(0) = f
1
(0) > g(0) and f(1) = f
2
(0) > g(1). Let l
1
= minl :
f
1
(l) > f
2
(1) and put f(2) = f
1
(l
1
). Now f(2) > g(2) since f
2
(1) > g(2). Let
l
2
= minl : f
2
(l) > f
1
(l
1
+ 1) and let f(3) = f
2
(l
2
) > g(3) since f
1
(l
1
+ 1) >
g(3). And so on . . . .
In general dene the sequence l
n
: n ) as l
0
= 0 and
l
2n+1
= minl : f
1
(l) > f
2
(l
2n
+ 1)
and
l
2n+2
= minl : f
2
(l) > f
1
(l
2n+1
+ 1).
Let
f(n + 1) =
_
f
1
(l
n
) if n is even
f
2
(l
n
) if n is odd
.
It is clear that f M. Easy induction shows that f dominates g. Contradic-
tion.
To get the sequence the desired sequence k
n
: n ) take a model M [=
ZFC

containing H
n
: n ) and F

: < . Since < add(c, ^) d


4
4
3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


we can assume that [M[ < d. Apply the above lemma to get a function g and
dene k
n
= g(n) for n . It is clear that this is the sequence we are looking
for.
Now dene for n ,

I
n
= [k
2n1
, k
2n+1
]
and
T
n
=
_
s 2

In
: j [k
2n
, k
2n+1
] t H
j
s

I
n
= t

I
n
_
.
Note that for every n,
[T
n
[
2
|

In|
2
kn

k2n+1

j=k2n
[H
j
[
2
j

1
2
n
.
To nish the proof x < and k . By the construction there exists
n > k and m such that
k
2n
< k

m
< k

m+1
< k
2n+1
.
Suppose that s F

I
n
. Then there exists x F
k

such that s x. Further-


more, there exists j [k

m
, k

m+1
) such that xj H
j
. It follows that s T
n
.
Now we will prove another combinatorial lemma describing the structure of
closed measure zero sets.
Let I
n
: n be a partition of into disjoint intervals such that [I
n
[ > n.
For n < m let
Seq
n,m
= s : dom(s) [n, m] & j dom(s) s(j) I
j
.
For every s Seq
n,m
dene
C
s
=
_
_
_
t : dom(t) =
m
_
j=n
I
j
& j dom(s) t
_
s(j)
_
= 0
_
_
_
.
For k, j let
C
j
k
=
_ _
t 2
Ij
: t(k) = 0
_
if k I
j
2
Ij
otherwise
.
Note that we can identify the set C
s
with

m
j=n
C
j
s(j)
in the following way:
t C
s
t
n
, t
n+1
, . . . , t
m
)
m

j=n
C
j
s(j)
t = t
n

t
n+1

t
m
.
Fix n < m and let I = I
n
I
n+1
I
m
. Suppose that T 2
I
is a nite
tree such that
5
4
3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


1. s T t T (s t & [t[ = [I[),
2. m(T)
1
4
.
Lemma 2.4 Suppose that for some s Seq
n,m
, C
s
=

m
j=n
C
j
s(j)
T. Then
there exists k [n, m) and t T

k1
j=1
C
j
s(j)
(if k = n then t = ) such that
t

C
k
s(k)
m
_
T[t

]
_
>
_
1 +
1
2
k
_
m
_
T[t]
_
.
Proof Suppose not. We build by induction a sequence t
j
: j [n, m1])
such that t
j
C
j
s(j)
and m
_
T[t
j

t
j+1
]
_
(1 + 2
j
) m
_
T[t
j
]
_
for j < m.
After m1 many steps we get that
m
_
T[t
n

t
n+1

t
m1
]
_
m(T)
m1

j=n
_
1 +
1
2
j
_
<
1
2
.
Therefore there is t
m
C
m
s(m)
T[t
n

t
n+1

t
m1
]. This is a contradiction
since
t = t
n

t
n+1

t
m
C
s
T.
Suppose that t T and [t[ = [

k
j=n
I
j
[ for some k [n, m). Let
S
k+1
t
=
_
l I
k+1
: t

C
k+1
l
m
_
T[t

]
_
>
_
1 +
1
2
k
_
m
_
T[t]
_
_
.
Note that the sets
_
C
k+1
l
: l I
k+1
_
are independent. Therefore the set
_
lS
k+1
t
_
t

C
k+1
l
T[t

]
has measure at least
_
1 2
|S
k+1
t
|
_

_
1 +
1
2
k
_
m
_
T[t]
_
.
Since this set is included in T[t] we get
_
1 2
|S
k+1
t
|
_

_
1 +
1
2
k
_
1.
Therefore
[S
k+1
t
[ k + 1.
6
4
3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


Let S
k+1
=
_
l I
k+1
: t T l S
k+1
t
_
. Then
[S
k+1
[ (k + 1)
k

j=n
2
|Ij|
.
Also if t = then dene
S
n

=
_
l I
n
: t

C
n
l
m
_
T[t

]
_
>
_
1 +
1
2
n
_
m(T)
_
.
Similarly we get [S
n

[ n + 1.
Note that in particular we get that the size of S
k
does not depend on the
size of I
k
.
Combining 2.4 with the observations above we get the following:
Lemma 2.5 Suppose that I = I
n
I
n+1
I
m
and T 2
I
such that
m(T) <
1
4
. Then there exists a sequence S
k
: k [n, m]) such that
1. S
k
I
k
for k [n, m],
2. [S
k
[ (k + 1)

k1
j=n
2
|Ij|
for k (n, m] and [S
n
[ n + 1,
3. for every s Seq
n,m
, if C
s
T then there exists k [n, m] such that
s(k) S
k
.
We conclude this section with a theorem of Miller which gives an upper
bound for cov(c, ^). We will prove it here for completeness.
Theorem 2.6 (Miller [Mi]) add(c, ^) d and cof (c, ^) b.
Proof Suppose that H 2

is a measure zero set. Using 2.2, we can nd


a sequence H
n
: n ) such that H
n
2
n
,

n=1
[H
n
[ 2
n

1
4
and
H x 2

n xn H
n
.
Dene for n ,
f
H
(n) = min
_
_
_
m :

j=m
[H
j
[
2
j
<
1
4
n
_
_
_
.
Suppose that f

is an increasing function. Let


G
f
= x 2

: n x
_
f(n)
_
= 0.
Clearly G
f
is a closed measure zero set.
Lemma 2.7 If f
H

f then G
f
, H.
7
4
3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


Proof Suppose that f
H

f. Without loss of generality we can assume


that f
H
(n) < f(n) for all n. For n dene

H
n
=
_
s 2
fH(n+1)
: j
_
f
H
(n), f
H
(n + 1)
_
t H
j
sj = t
_
.
Note that for all n,
[

H
n
] =
fH(n+1)
_
j=fH(n)
[H
j
] and m(

H
n
) 4
n
.
By compactness, if G
f
H then for some n,
G
f

fF (n+1)
_
j=1
[H
j
] =
_
jn
[

H
j
].
We will show that this inclusion fails for every n which will give a contradiction.
Fix n . Note that it is enough to nd s 2
fH(n+1)
such that s
_
f(j)
_
= 0
and sf
H
(j + 1) ,

H
j
for j n.
We will use the following simple construction.
Lemma 2.8 Suppose that n
1
< n
2
< n
3
and that T 2
[n1,n3]
is such that
m(T) = a <
1
2
. For l [n
2
, n
3
] let C
l
=
_
s 2
[n2,n3]
: s(l) = 0
_
. Then for ev-
ery l [n
2
, n
3
] there exists s C
l
such that the set T[s] =
_
t 2
[n1,n2)
: t

s T
_
has measure 2a.
Proof Fix l [n
2
, m
3
] and choose s such that m(T[s]) is minimal.
If T[s] = we are done. Otherwise
m(T)
1
2
m(T[s]).
It follows that m(T[s]) 2a.
We will build by induction sequences s
n
, s
n1
, . . . , s
0
and sets H

n
, H

n1
, . . . , H

0
such that for all j n,
1. dom(s
j
) = [f
H
(j), f
H
(j + 1)),
2. H

j
2
fH(j+1)
,
3. m(H

j
[s
j
]) 2 m(H

j
).
Let H

n
=

H
n
and let s
n
2
[fH(n),fH(n+1))
be the sequence obtained by applying
2.8 to H

n
and C
f(n)
.
Suppose that H

nj
and s
nj
are already constructed. Let
H

nj1
=

H
nj1
H

nj
[s
nj
]
8
4
3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


and let s
nj1
be the sequence obtained by applying 2.8 to H

nj1
and C
f(nj1)
.
Let s = s
0

s
1

s
n
. Note that s
_
f(j)
_
= 0 for all j n. We have to
check that sf
H
(j + 1) ,

H
j
for j n. Suppose this is not true. Pick minimal
j such that
sf
H
(j + 1) = s
0

s
1

s
j


H
j
.
By the choice of s
j
we have
s
0

s
1

s
j1


H
j1


H
j
[s
j
].
Since j was minimal,
s
0

s
1

s
j1


H
j
[s
j
].
Proceding like that we get that
s
0

s
1

s
j2


H
j
[s
j
][s
j1
]
Finally
s
0


H
j
[s
j
][s
j1
] [s
1
] H

0
which is a contradiction.
Now we are ready to nish the proof of the theorem. Suppose that F

is a dominating family which consists of increasing functions. Consider the set

fF
G
f
. We claim that this set does not have measure zero. It follows from the
fact that if H is a measure zero set then there exists f F such that f
H

f.
In particular G
f
, H.
Similarly, if B ^ is a family of size < b then there exists f

such that
H B f
H

f.
Thus G
f
, H for any H B.
3 Cohen reals from closed measure zero sets
The goal of this section is to prove that add(c, ^) = cov(/). In fact we have
the following:
Theorem 3.1
1. add(c, ^) = cov(/). In particular add(c) = add(/),
2. cof (c, ^) = unif (/). In particular cof (c) = cof (/).
9
4
3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


Proof Note that by 1.1 and 2.6, we get
add(/) = mincov(/), b add(c, ^) b.
Therefore the equality add(c) = add(/) follows from the inequality add(c, ^)
cov(/).
Similarly, to show that cof (c) = cof (/) we have to check that cof (c, ^)
unif (/).
(1) add(c, ^) cov(/).
By the rst part of 1.2, it is enough to prove that for every family F

of size < add(c, ^) there exists a function g

such that
f F

n f(n) = g(n).
Fix a family F as above.
For every f F let
f

(n) = max f(i) : i n + 1 for n .


We will need two increasing sequences m
n
, l
n
: n such that
1. m
0
= l
0
= 0,
2. l
n+1
= l
n
+ 2
mn
(n + 1),
3. f F

n m
n+1
> f

(l
n+1
)
ln+1
+m
n
.
The existence of these sequences follows from the fact that [F[ < d.
Let I
n
= [m
n
, m
n+1
) and J
n
= [l
n
, l
n+1
) for n . Without loss of gener-
ality we can assume that [I
n
[ = K
n
|Jn|
for some K
n
. Thus we can identify
elements of I
n
with K
n
Jn
.
For every f F and n dene

f(n) = fJ
n
. By the choice of sequences
I
n
, J
n
: n ) we have
f F

n

f(n) I
n
.
Using the notation from previous section, dene for f F,
C
f
=

n
C

fn
.
Note that the sets C
f
are closed sets of measure zero.
Since [F[ < add(c, ^), the set

fF
C
f
has measure zero.
By 2.1, there exist sequences

I
n
, T
n
: n ) such that for all n, T
n
2

In
,
[T
n
[ 2
|

In|
2
n
and
f F

n C
f

I
n
T
n
.
10
4
3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


Moreover, without loss of generality we can assume that whenever I
m


I
n
,=
then I
m


I
n
for n, m .
We will build the function g

we are looking for from the sequences


T
n
: n ) and I
n
: n ).
For every n let v
n
be such that

I
n
= I
vn
I
vn+1
I
vn+11
.
Note that for f F and n ,
C
f

I
n
= C

f[vn,vn+1)
.
Now we are ready to dene function g. For every n we will dene g

I
n
using
the set T
n
.
Fix n and consider the set T
n
2

In
. By 2.5 there exists a sequence
S
k
: k [v
n
, v
n+1
)) such that
1. S
k
I
k
for k [v
n
, v
n+1
),
2. [S
k
[ (k + 1)

k1
j=n
2
|Ij|
for k (v
n
, v
n+1
) and [S
vn
[ n + 1,
3. for every s Seq
vn,vn+11
, if C
s
T then there exists k [v
n
, v
n+1
) such
that s(k) S
k
.
Note that for every k [v
n
, v
n+1
),
[S
k
[ (k + 1)
k1

j=n
2
|Ij|
(k + 1)
k1

j=n
2
mj+1mj
(k + 1) 2
mj
[J
k
[.
We can view S
k
as a subset of K
J
k
k
of size [J
k
[. For k [v
n
, v
n+1
) let
s
k
K
J
k
k
be such that
t S
k
l J
k
s
k
(l) = t(l).
Dene
g

I
n
= s
vn

s
vn+11
.
Note that g

I
n
diagonalizes all sets S
k
for k [v
n
, v
n+1
).
Now we are ready to nish the proof. Suppose that f F. Therefore there
exists innitely many n such that
C
f

I
n
= C

f[vn,vn+1)
T
n
.
In particular there exists k [v
n
, v
n+1
) such that

f(k) = fJ
k
S
k
. Thus
there exists j J
k
such that
f(j) = s
k
(j) = g(j)
11
4
3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


which nishes the proof of the rst part of the theorem. Note that we only used
the the fact that m(T
n
)
1
4
for n .
(2) unif (/) cof (c, /).
To prove this inequality we have to dualize the above argument. Suppose
that B ^ is a family of size witnessing that cof (c, ^) = . We will
construct a family F

of size such that


f

g F

n f(n) = g(n).
By 1.2, this will nish the proof.
Since cof (c, ^) b we can nd a family G

of size which is un-


bounded and consists of increasing functions.
Let G = f

: < and B = H

: < . Without loss of generality we


can assume that
H

= x 2

n xn H

where

n=1
[H

n
[ 2
n
< . For every , < and n dene

I
,
n
= [f

(2n 1), f

(2n + 1)]
and
T
,
n
=
_
s

I
,
n
: j [f

(2n), f

(2n + 1)] t H

j
s

I
,
n
= t

I
,
n
_
Let
W =
_
, ) : n [T
,
n
[ 2
|

I
,
n
|
2
n
_
.
Arguing as in the proof of 2.1, we show that for every closed measure zero
set F 2

there exists , ) W such that

n F

I
,
n
T
,
n
.
Let V be the set of triples , , )
3
such that , ) W and the
partition
_
f

(n), f

(n + 1)
_
: n ) is ner that

I
,
n
: n ).
For every triple , , ) V let g
,,

be the function g dened in the


proof above.
Let
F =
_
g
,,
: , , ) V
_
.
We will show that this family has required properties. Suppose that f

.
Find , < such that
1. f

(n + 1) f

(n) + 2
f(n)
(n + 1),
2.

n f

(n + 1) > f

_
f

(n + 1)
_
f

(n+1)
+f

(n)
12
4
3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


where f

(n) = maxf(1), . . . , f(n) + 1.


Dene I
n
=
_
f

(n), f

(n + 1)
_
and J
n
=
_
f

(n), f

(n + 1)
_
for n . As in
the above part we have

n

f(n) I
n
.
Now we can nd , ) W such that

n C
f

I
,
n
T
,
n
.
It follows that

n f(n) = g
,,
(n)
which nishes the proof.
We conclude this section with two applications.
In [Mi1] it is proved that:
Theorem 3.2 (Miller) add(^) b and cof (^) d.
Theorem 3.3 (Bartoszynski, Raisonnier, Stern [Ba], [RS]) add(^) add(/)
and cof (^) cof (/).
Proof We have
add(^) minb, add(c, ^) = minb, cov(/) = add(/).
Similarly
cof (^) maxd, cof (c, ^) = maxd, unif (/) = cof (/).
Also we get another proof of the main result from [BJ]:
Theorem 3.4 (Bartoszynski, Judah) cf
_
cov(/)
_
add(^).
Proof Clearly cf
_
add(c, ^)
_
add(^).
4 Cardinals cov(E) and unif(E)
In this section we will prove some results concerning covering number of c. Most
of the results are implicite in [Ba2] and [BJ1].
Let us start with the following easy observation.
Lemma 4.1
1. Every null set can be covered by d many closed null sets,
2. Every null set of size < b can be covered by a null set of type F

.
13
4
3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


Proof Suppose that G is a null subset of 2

. As in 2.2, we can assume that


G = x 2

n xn F
n

where

n=1
[F
n
[ 2
n
< . For every x G let f
x

be an increasing
enumeration of the set n : xn F
n
. For a strictly increasing function
f

let
G
f
= x 2

n m [n, f(n)] xm F
m
.
It is clear that for every f

the set G
f
G is a measure zero set of type
F

. Notice also that if f


x

f then x G
f
.
(1) Let F

be a dominating family of size d which consists of increasing


functions. Then by the above remarks
G =
_
fF
G
f
.
(2) Suppose that X G is a set of size < b. Let f be an increasing function
which dominates all functions f
x
: x X. Then X G
f
.
As a corollary we get:
Theorem 4.2
1. If cov(/) = d then cov(c) = max cov(/), cov(^),
2. If unif (^) = b then unif (c) = min unif (/), unif (^).
Proof Since c / ^ we have
cov(c) max cov(/), cov(^)
and
unif(c) min unif(/), unif (^) .
By the previous lemma
max cov(/), d cov(c)
and
unif (c) min unif (^), b
which nishes the proof.
Suppose that f

and

n=1
2
f(n)
< . Dene

f

_
[]
<
_

& n
_
(n) 2
f(n)
&
[(n)[
2
f(n)

1
4
n
_
14
4
3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


and

f

_
[]
<
_

& n (n) 2
f(n)
&

n
[(n)[
2
f(n)

1
4
n
and let A
f
=

n=1
2
f(n)
.
Notice that
f

f
.
For
f

f
dene dene set H

as follows:
Let k
n
= 1+2+ +f(n) for n . Identify natural numbers 2
f(n)
with
0-1 sequences of length f(n) and dene
H

= x 2

n x[k
n
, k
n+1
) (n) .
Note that
(H

n=m
(x 2

: x[k
n
, k
n+1
) (n))

m=1

n=m
[(n)[
2
f(n)
= 0.
For x 2

. Dene h
x
(n) = x[k
n
, k
n+1
) for n . Clearly h
x
corresponds
to an element of A
f
.
Finally we have
x H

n h
x
(n) (n).
Theorem 4.3 Suppose that C c. Then there exists f

and
f
such
that C H

.
Proof Suppose that C 2

is a null set of type F

. Represent C as

n
C
n
where C
n
: n ) is an increasing family of closed sets of measure
zero. Dene sequence k
n
: n ) as follows: k
0
= 0 and
k
n+1
= min
_
m > k
n
: T
n
2
m
_
C
n
[T
n
] &
[T
n
[
2
m

1
4
kn
__
.
Let I
n
= [k
n
, k
n+1
) and J
n
= sI
n
: s T
n
for n . We can see that for all
n
[J
n
[
2
|In|
2
kn

1
4
kn

1
2
n
.
We also have
F x 2

n xI
n
J
n
= H

where f(n) = [I
n
[ and (n) = J
n
for all n. By the above remarks
f
.
For an increasing function g

dene g

as g

(0) = 0 and g

(n+1) =
g(g

(n) + 1).
Lemma 4.4 Suppose that f, g

are increasing functions and


f
.
1. If f

g then there exists


g
such that H

,
15
4
3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


2. if g ,

f then there exists


g
such that H

.
Proof Let I
n
=
_
f(n), f(n + 1)
_
and I

n
=
_
g

(n), g

(n + 1)
_
for n .
Note that if f

g then

n m I
m
I

n
and if g ,

f then

n m I
m
I

n
.
Dene
(n) =
_ _
s 2
I

n
: m
_
I
m
I

n
& sI
m
(m)
__
if m I
m
I

n
2
I

n
otherwise
It follows that
g
in the rst case and
g
in the second case.
Moreover, the inclusion, H

is an immediate consequence of the above


denition.
As a consequence we get:
Theorem 4.5 Suppose that F

: < is a family of elements of c.


1. If < b then there exists a function g

and a family

: <
g
such that F

for < ,
2. if < d then there exists a function g

and a family

: <
g
such that F

for < .
The following fact follows immediately from 4.5.
Theorem 4.6 If cov(c) < d then there exists f

such that cov(c) is equal


to the size of the smallest family
f
such that
h A
f

n h(n) (n).
As an corollary we get the following:
Theorem 4.7 (Miller) If cov(c) d then cf
_
cov(c)
_
>
0
.
Proof Suppose that cf(cov(c)) =
0
. Since d has uncountable cardinality
we have cov(c) < d. By 4.6 under this assumptions there exists g

such
that cov(c) is the size of the smallest family
g
such that
h A
g

n h(n) (n).
Assume that is the smallest family having above properties and let
n
: n
be an increasing family such that =

n

n
and [
n
[ < [[ for all n .
16
4
3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


By the assumption for every m there exists a function h
m
A
g
such
that
m
m

n h
m
(n) , (n).
For dene k

0
= 0 and for n
k

n+1
= min
_
m > k

n
: j n i [k

n
, m) h
j
(i) , (i)
_
.
Since [[ < d we can nd an increasing function r

such that

n k

n
r(n).
Let h = h
1
[r

(0), r

(1))

h
2
[r

(1), r

(2))

h
3
[r

(2), r

(3))

. . . .
Fix . By the assumption about r we have

n m > n r

(n) < k

m
< k

m+1
< r

(n + 1).
But this means that
i [r(n), r(n + 1)) h
n+1
(i) = h(i) , (i).
Since is an arbitrary element of it nishes the proof.
5 Consistency results
The goal of this section is to show that cov(c) > max cov(^), cov(/) and
unif (c) < min unif (^), unif (/) are both consistent with ZFC. We use the
technique developed in [JS].
Lemma 5.1 Suppose that T is a notion of forcing satisfying ccc. Let

C be a
T-name for an element of c.
1. If T does not add dominating reals then there exists f

V and a
T-name such that |
P

f
and |
P

C H

,
2. if T is

-bounding then there exists f

V and a T-name such


that |
P

f
and |
P

C H

.
Proof Follows immediately from 4.4.
Denition 5.2 Suppose that N [=ZFC

. A function x 2

is called N-big i
x ,
_
(c N).
We say that a partial ordering T satisfying ccc is good if for every model
N H() and every lter G which is T-generic over V, if x 2

is N-big
then x is N[G]-big.
17
4
3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


Let B denote the random real forcing.
Theorem 5.3 B is good.
Proof Suppose that x is N-big. Let

C N be a B-name for an element
of c. Since B is

-bounding, by 5.1, we can nd a function f

N and a
B-name N for an element of
f
such that |
B

C H

.
For s 2
f(n)
dene B
n,s
= [[s (n)]]
B
. Let
(n) =
_
s : (B
n,s
)
1
2
n
_
for n .
Note that since
|
B
[ (n)[
2
f(n)

1
4
n
we get that
[(n)[
2
f(n)

1
2
n
for n .
Suppose that p |
B
n m xn (n). Find k such that (p) 2
k
.
Since x is N-big there exists n k such that s = xI
n
, (n). In particular
(B
n, s
) < 2
k
. Let q = p B
n, s
. It is clear that
q |
B
xI
n
, (n)
which gives a contradiction.
Lemma 5.4
1. If T and Q are good forcing notions then T Q is good.
2. If
_
T

,

Q

: <
_
is a nite support iteration such that
(a) |

is good,
(b) |

V is unbounded.
then T

= lim
<
T

is good.
Proof The rst part is obvious. We will prove the second part by induction
on . Without loss of generality we can assume that is a limit ordinal. Suppose
that the lemma is true for < . Let N H() be a model and let

C be a
T

-name for an element of c N. It is well known that under the assumptions


T

does not add dominating reals. Therefore there exists f

N and a
T

-name for an element of


f
such that
|


C H

.
18
4
3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


Assume that x is N-big and suppose that for some p T

,
p |

n > n
0
x[f(n), f(n + 1)) (n).
Dene a sequence p
n
: n ), k
n
: n ) N and
f
such that
1. p = p
0
p
1
p
2
. . .,
2. p
n+1
|

j k
n
(j) = (j),
3. p
n+1
|

j [k
n
, k
n+1
] [(j)[ 2
f(j)
4
j
.
Since x is N-big there exists m > n
0
such that x[f(m), f(m + 1)) , (m).
Therefore p
m
| x[f(m), f(m+ 1)) , (m). In particular,
p ,| x[f(m), f(m+ 1)) , (m)
which is a contradiction.
Theorem 5.5 It is consistent with ZFC that
unif(c) < min unif(^), unif (/) .
Proof Let T
2
be a nite support iteration of length
2
of random real
forcing. Let G be a T
2
-generic lter over a model V [= GCH. Since T
2
adds
random and Cohen reals we have V[G] [= unif (/) = unif (^) =
2
. We will
show that V[G] [= unif (c) =
1
. It is enough to show that V[G] [= 2

V , c.
Suppose that C V[G] c. Let

C be a T
2
-name for C. Let N H()
be a countable model containing

C and T
2
. Since N is countable there exists
x 2

V which is N-big. By 5.4, x is also N[G]-big. In particular x , C.


Theorem 5.6 It is consistent with ZFC that cov(c) > max cov(^), cov(/).
Proof Let T
1
be a nite support iteration of length
1
of random real
forcing. Let G be a T
1
-generic lter over a model V [= cov(c) =
2
.
It is clear that V[G] [= cov(^) = cov(/) =
1
. We will show that
V[G] [= cov(c) =
2
.
Suppose that C

: <
1
V[G] c. Let

C

be a T
1
-name for C

.
Let N H() be a model of size
1
containing all names

C

and T
1
. Since
V [= cov(c) >
1
there exists x 2

V which is N-big. By 5.4, x is also


N[G]-big. In particular, x ,

<1
C

.
19
4
3
9


r
e
v
i
s
i
o
n
:
1
9
9
9
-
0
4
-
1
4







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


References
[Ba] T. Bartoszynski Additivity of measure implies additivity of category,
Transactions of AMS vol. 1, 1984
[Ba1] T. Bartoszynski Combinatorial aspects of measure and category , Fun-
damenta Mathematicae, 1987
[Ba2] T.Bartoszynski On covering of the real line by null sets, Pacic Jour-
nal of Mathematics, vol. 1, 1988
[BJ] T.Bartoszynski, H.Judah On the conality of the smallest covering of
the real line by meager sets, Journal of Symb. Logic vol. 54, no.3,
1989
[BJ1] T.Bartoszynski, H.Judah Jumping with random reals, Annals of Pure
and Applied Logic, vol.48, 1990
[FM] D. Fremlin, A. Miller On some properties of Hurewicz Menger and
Rothberger , Fundamenta Mathematicae, vol.129, 1988
[JS] H.Judah, S.Shelah The KunenMiller chart, Journal of Symb. Logic
vol. 55, no.3 1990
[Mi] A. Miller Some properties of measure and category Trans. AMS vol.
266, 1981
[Mi1] A. Miller Additivity of measure implies dominating reals Proceedings
of AMS vol. 91, 1984
[O] J. Oxtoby Measure and category, Springer Verlag
[RS] J.Raisonnier, J.Stern The strength of measurability hypotheses, Israel
Journal of Mathematics vol. 50 1985
20

Das könnte Ihnen auch gefallen