Anda di halaman 1dari 17

8

0
9


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
7
-
3
1







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


COMPARING THE UNIFORMITY INVARIANTS OF NULL SETS FOR
DIFFERENT MEASURES
SAHARON SHELAH AND JURIS STEPR

ANS
Abstract. It is shown to be consistent with set theory that the uniformity invariant for Lebesgue
measure is strictly greater than the corresponding invariant for Hausdor r-dimensional measure where
0 < r < 1.
Contents
1. Introduction 1
2. Notation 2
3. Continuous mappings between Lebesgue and Hausdor measures 3
4. A preliminary norm 9
5. Finding nite sets with large norm 11
6. The forcing partial order 13
7. Remarks and open questions 16
References 16
1. Introduction
The uniformity invariant for Lebesgue measure is dened to be the least cardinal of a non-measurable
set of reals, or, equivalently, the least cardinal of a set of reals which is not Lebesgue null. This has
been studied intensively for the past 30 years and much of what is known can be found in [?] and other
standard sources. Among the well known results about this cardinal invariant of the continuum is that
it can equally well be dened using Lebesgue measure on R
n
without changing the value of the cardinal.
Indeed, equivalent denitions will result by using any Borel probability measure on any Polish space.
However, the question of the values of uniformity invariants for other, non--nite Borel measures is not
so easily answered. This paper will deal with the most familiar class of such measures, the Hausdor
measures for fractional dimension. Observe that by the previous remarks, the least cardinal of any
non-measurable subset of any -nite set will be the same as the uniformity invariant for Lebesgue
measure. In other words, this paper will be concerned with the uniformity invariant of the ideal of
-nite sets with respect to a Hausdor measure.
1991 Mathematics Subject Classication. 03E17, 28A78.
Key words and phrases. Lebesgue measure, Hausdor measure, null set, proper forcing, product forcing, uniformity
cardinal invariant.
The rst authors research for this paper was supported by the United States-Israel Binational Science Foundation
and by grant No. NSF-DMS97-04477 of the National Science Foundation of the United States. This is number 809 in
his personal numbering system. The second authors research for this paper was supported by the Natural Sciences and
Engineering Research Council of Canada and was completed while enjoying the hospitality of Rutgers University. The
second author expresses his thanks to D. Fremlin and A. Miller for enduring a lengthy series of lectures based on an early
version of this paper, a version even more awkward than the current.
1
8
0
9


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
7
-
3
1







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


2 S. SHELAH AND J. STEPR

ANS
It will be shown that given any real number r in the interval (0, 1) it is consistent with set theory that
every set of reals of size
1
is Lebesgue measurable yet there is a set of reals of size
1
which is not a
null set with respect to r-dimensional Hausdor measure. This answers Question FQ from D. Fremlins
list of open questions [1]. However, the motivation was an attempt to resolve the following question
posed by P. Komjath.
Question 1.1. Suppose that every set of size
1
has Lebesgue measure zero. Does it follow that the
union of any set of
1
lines in the plane has Lebesgue measure zero?
It is worth noting that this is really a geometric question since [7] provides a negative answer to the
version of the problem in which lines are replaced by their topological and measure theoretic equivalents.
To see the relationship between Question 1.1 and the topic of this paper consider that it is easy to nd
countably many unit squares in the plane such that each line passes through either the top and bottom
or the left and right sides of at least one of these squares. It is therefore possible to focus attention on
all lines which pass through the top and bottom of the unit square. For any such line L there is a pair
(a, b) such that both the points (a, 0) and (1, b) belong to L. If the mapping which sends a line L to
this pair (a, b) is denoted by then it is easy to see that is continuous and that if S [0, 1]
2
is a
square of side then the union of
1
S has measure while S itself has measure
2
. In other words,
the Lebesgue measure of the union of the lines belonging to
1
X is no larger than the 1-dimensional
Hausdor measure of X for any X [0, 1]
2
. In other words, a negative answer to Question 1.1 would
imply that there is X [0, 1]
2
of size
1
which is not null with respect to linear Hausdor measure even
though every set of reals of size
1
is null. The consistency of this will be a consequence of Corollary 6.2.
The proof will rely partially on arguments from [5] and [6] in which a single stage of a forcing iteration
that would achieve the desired model was described. The material in 3 and 4 is a reorganized and
simplied version of 4, 5 and 6 of [6] which has been suitably modied for the current context. The
approach taken here diers from the earlier attempt in that the forcing used is nite branching rather
than innite branching as in [6] and this allows the use of product forcing along lines similar to those in
[2]. The new ingredient needed is described in 5. The arguments presented in 6 use ideas explained
in greater detail in [4], however familiarity with that paper is not be required in order to follow the
reasoning presented here.
The logical requirements guiding the organization of the paper may not have resulted in optimal
organization for the purposes of comprehension. Some readers may prefer to start with 6 after having
read 2, Denition 3.2 and Denition 4.2.
2. Notation
If X R and r (0, 1) then the inmum of all

i=0
(b
i
a
i
)
r
where (a
i
, b
i
)

i=0
is a cover of X by
intervals of length less than is often denoted by H
r

(X). The r-dimensional Hausdor capacity of X is


denoted by H
r

(X) and is dened to be the inmum of all

i=0
(b
i
a
i
)
r
where (a
i
, b
i
)

i=0
is a cover
of X by arbitrary intervals. The r-dimensional Hausdor measure of a set X is denoted by H
r
(X) and,
when dened, is equal to its outer measure lim
0
H
r

(X). Since it can be shown


1
that H
r

(X) = 0 if
and only if H
r
(X) = 0, in order to establish the main result it suces to show that it is consistent with
set theory that every set of reals of size
1
is a Lebesgue null set yet there is a set of reals of size
1
which
is not a null set with respect to r-dimensional Hausdor capacity. Actually, it will be more convenient
to work with measures and capacities on the Cantor set and to replace intervals by dyadic intervals.
This amounts to dealing with net measures as described in [3]. Since r-dimensional net measures have
the same null sets as r-dimensional Hausdor measures this will not be of signicance to the results of
this paper.
1
See Lemma 4.6 in [3] for example.
8
0
9


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
7
-
3
1







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


UNIFORMITY INVARIANTS 3
Notation 2.1. For the rest of this paper let r be a xed real number such that 0 < r < 1.
Notation 2.2. Let C denote 2
N
with the usual product topology. Let Q be the tree f n : f
C and n N. For a Q let [a] = x C : a x and for A Q let [A] =

aA
[a]. Let
C

=
_
A Q :

aA
2
|a|r

_
and let C

= A C

: [A[ <
0
. For consistency of notation, let [Q]
<
0
be denoted by C

. Let
C
1

= A Q :

tA
2
|t|
. For X C dene
r

(X) = inf : (A C

) X [A]. The usual


product measure on C will be denoted by .
Notation 2.3. Some notation concerning trees will be established. By a sequence will always be meant
a function f : n X where n N and X is some set. Sequences will occasionally be denoted as n-tuples
(x
1
, x
2
, . . . , x
n
) and, in particular, singleton sequences will be denoted by (x). If t and s are sequences
the concatenation of s followed by t will be denoted by (s, t) . This is consistent with considering X
k
to be a set of sequences because if x X
k
and y Y
m
then (x, y) X
k
Y
m
. If T is a tree then T is
a set of sequences closed under restriction to initial segments. If t T then o
T
(t) = x : (t, (x)) T.
Furthermore, Tt) = s T : t s or s t. If m N then T[m] = t T : [t[ = m and
T[< m] = t T : [t[ < m and T[ m] = t T : [t[ m.
Notation 2.4. For any X C
m
and z C
k
the set x C
mk
: (z, x) X will be denoted by
X(z, ). Similarly, if F : C
m
X is a function and z C
k
then F(z, ) will represent the function
F(z, ) : C
mk
X dened by F(z, )(x) = F(z, x).
3. Continuous mappings between Lebesgue and Hausdorff measures
The goal of this section is to exploit the dierence between Lebesgue measure and r-dimensional
Hausdor measure. It will be shown that for any continuous function from the reals to the reals there
are arbitrary small sets in the sense of Lebesgue measure whose pre-image is as large as desired in the
sense of r-dimensional Hausdor measure. For reasons which will reveal themselves in 4 it will also be
necessary to consider similar results for products of the reals.
Denition 3.1. If > 0 is a real number and X C dene

(X) by

(X) = inf
r

(X Z) : Z C and (Z) < .


Lemma 3.1. If 1 > > 0 and > 0 then there is > 0 such that for all but nitely many m N, any
measurable E C and any measurable D E

(D)

(E)
provided that (D [s]) (E [s]) for each s Q[m].
Proof. Let > 0 be suciently small that

2
>
4

and then let m N be so large that the inequality


2
m(1r)

r
4
r
_

2

4

_
>

(E)
is satised.
8
0
9


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
7
-
3
1







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


4 S. SHELAH AND J. STEPR

ANS
Suppose that (Z) < and A Q are such that D Z [A] and

aA
2
|a|r
<

(E).
Let A

= Q[ m] A and let E

= E [A

]. Let B
0
= b Q[m] : [b] [A

] = and note that


E

[B
0
] and that if b B
0
then E [b] = E

[b]. Now let


B
1
=
_
b B
0
: (E [b]) >
1
2
(E

)
2
m
_
.
Notice that (E

) because

aA

2
|a|r
<

(E) and E E

[A

]. It follows from a Fubini


argument that
(3.1) [B
1
[
2
m
(E

)
2 (E

)

2
m

2
.
Next, let
B
2
=
_
b B
1
: (Z [b]) >
(E [b])
2
_
and in order to see that
(3.2) [B
2
[
2
m+2

assume the opposite. Then the following sequence of inequalities


> (Z)

bB
2
(Z [b])

2

bB
2
(E [b])

bB
2
(E

)
2
m+2
[B
2
[

2
m+2

yields a contradiction. (The fourth inequality uses that B
2
B
1
.)
Let B
3
= B
1
B
2
and observe that it follows from Inequalities 3.1 and 3.2 that
(3.3) B
3

2
m

2

2
m+2

= 2
m
_

2

4

_
.
It follows that

aA
2
|a|r

bB
3
_
_

aA\A

,ab
2
|a|r
_
_

bB
3
((D [b] Z))
r

bB
3
((D [b]) (Z [b]))
r
.
Since B
3
B
2
= it follows, using the hypothesis on D, that the last term dominates

bB
3
((E [b]) (/2)(E [b]))
r


r
2
r

bB
3
((E [b]))
r


r
2
r

bB
3
_
(E

)
2
m+1
_
r

r
4
r
[B
3
[
_
(E

)
2
m
_
r

2
m(1r)

r
4
r
_

2

4

_
and hence

aA
2
|a|r
>

(E) which is impossible.


If X C then F : X C will be said to have small bres if and only if (F
1
x) = 0 for each
x C. The proof of Theorem 3.1 and the lemmas preceding it will rely on decomposing an arbitrary
continuous function into a piece that has small bres and a piece which has countable range.
8
0
9


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
7
-
3
1







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


UNIFORMITY INVARIANTS 5
Lemma 3.2. Let 0 < < 1 and suppose that X
s

sQ
is an indexed family of mutually independent
0, 1-valued random variables, each with mean . Suppose that C C is a measurable set and that
F
j
: C C is a measurable function with small bres for 1 j n. For any > 0 for all but nitely
many m N the probability that

_
_
n

j=1
_
_
_
sQ[m],Xs=1
F
1
j
[s]
_
_
_
_
>

n
(C)
2
is greater than 1 .
Proof. This is Lemma 3.1 in [5] or Lemma 6.2 in [6] except that it is stated here for C rather than
[0, 1].
Lemma 3.3. Suppose that
E C is a measurable set
T is a nite family of measurable functions with small bres from E to C
> 0, > 0.
Then there is > 0 such that for any > 0 and for any mutually independent, 0, 1-valued random
variables X
s

sQ
with mean and for all but nitely many m N the probability that the inequality
(3.4)

_
_

FF
F
1
_
_
_
sQ[m],Xs=1
[s]
_
_
_
_

(E)
holds is greater than 1 .
Proof. Let [T[ = n. Use Lemma 3.1 to choose > 0 and an integer k such that if D E is a measurable
set such that for each s Q[k]
(D [s])

n
2
(E [s])
then

(D)

(E). Let > 0. Now use Lemma 3.2 to conclude that for each t Q[k] for all but
nitely many m N and any mutually independent, 0, 1-valued random variables X
s

sQ[m]
with
mean , the probability that

_
_
[t]

FF
F
1
_
_
_
sQ[m],Xs=1
[s]
_
_
_
_


n
2
(E [t])
is greater than 1 2
k
. Hence the probability that this holds for all t Q[k] is greater than 1 and
so the hypothesis on k guarantees that Inequality 3.4 holds with at least the same probability.
Corollary 3.1. Suppose that C is a measurable subset of C
d+1
and T is a nite family of measurable
functions from C to C such that F(x, ) has small bres for each x C
d
and F T. If > 0 and
> 0 then there is > 0 such that for all > 0 there is some a C
1

such that the Lebesgue measure of


_
x C
d
:

FF
F(x, )
1
a
_

(C(x, ))
_
is at least 1 .
Proof. This is a standard application of Fubinis Theorem using Lemma 3.3 and the Law of Large
Numbers.
8
0
9


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
7
-
3
1







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


6 S. SHELAH AND J. STEPR

ANS
Lemma 3.4. Let E C be a measurable set and T a nite family of measurable functions from E to
C. Then for any > 0 and any > 0 there is > 0 and a C
1

such that

FF
F
1
a
_

(E).
Proof. Let [T[ = n. For each F T let Y
F
= F
1
y : (F
1
y) > 0 and let

F be dened by

F(z) =
_
F(z) if z E Y
F
z if z Y
F
.
Since each

F is measurable and has small bres it is possible to use Lemma 3.3 to conclude that there
is > 0 and an integer m and mutually independent, 0, 1-valued random variables X
s

sQ[m]
with
mean /3 such that the probability that

2
_
_

FF

F
1
_
_
_
sQ[m],Xs=1
[s]
_
_
_
_

(E)
is greater than 1/2.
Since the mean of each X
s
is /3 it is possible to choose m so large that the probability that

_
_
_
sQ[m],Xs=1
[s]
_
_
<

2
is also greater than 1/2. Hence there is a

C
1
/2
such that

2
_

FF

F
1
a

(E).
Now for each F T choose a nite set A
F
C such that (Y
F
F
1
A
F
) < /n. Then let a C
1

be
such that a

FF
A
F
a. It follows that

FF
F
1
a
_

FF

F
1
a
_
FF
(Y
F
F
1
A
F
)
_

(E)
because F
1
a

F
1
a (Y
F
F
1
A
F
) for each F T.
For the next denition recall Denition 3.1.
Denition 3.2. If : N R
+
, : N R
+
and X C
d
then the relation
,
(X) will be dened to
hold by induction on d. If d = 1 then
,
(X) if and only if
(0)
(X) (0) whereas, if d > 1, then

,
(X) holds if and only if
(d1)
(x C :
,
(X(x, ))) (d 1). Dene
+
(i) = (i +1) and

+
(i) = (i + 1).
The next lemma establishes that the top-down and bottom-up denitions of the relation
,
are
the same.
Lemma 3.5. Let d 2. For any : N R
+
, : N R
+
and X C
d+1
the relation
,
(X) holds
if and only if

+
,
+
_
z C
d
:
(0)
(X(z, )) (0)
_
.
8
0
9


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
7
-
3
1







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


UNIFORMITY INVARIANTS 7
Proof. Proceed by induction on d and observe that the case d = 2 is immediate from Denition 3.2.
Assuming the lemma established for d let X C
d+2
. Then the following sequence of equivalences
establishes the lemma:

,
(X)

(d+1)
__
z C
1
:
,
(X(z, ))
__
(d + 1)

(d+1)
__
z C
1
:

+
,
+
_
w C
d
:
(0)
(X(z, )(w, )) (0)
___
(d + 1)

+
(d)
__
z C
1
:

+
,
+
_
w C
d
:
(0)
(X((z, w), )) (0)
___

+
(d)

+
,
+
__
(z, w) C
1
C
d
:
(0)
(X((z, w), )) (0)
__

+
,
+
__
z C
d+1
:
(0)
(X(z, )) (0)
__

Lemma 3.6. If
,
(X) holds and X C
d
and < (i) for each i < d and A C
d
is such that
(A) <
d
then
,
(X A) holds where ()(i) = (i) .
Proof. Proceed by induction on d using Fubinis Theorem.
Theorem 3.1. Let : N R
+
and : N R
+
be functions such that (i) + (i) < 1 for each
i. Suppose that C is a closed subset of C
d
such that
,
(C) holds and that T is a nite family
of continuous functions from C to C. If > 0 then there is a C
1

and : N R
+
such that

,
_
FF
F
1
a
_
.
Proof. Proceed by induction on d, noting that if d = 1 then this follows from Lemma 3.4 by setting
= (0) in that lemma. So assume that the lemma has been established for d and that C is a closed
subset of C
d+1
,
,
(C) holds and T is a nite family of continuous functions from C to C and that
> 0. For each F T let
Y
F
= (x, y) C
d
C : (z C : F(x, y) = F(x, z)) > 0
and dene

F : C C by

F(x, y) =
_
F(x, y) if (x, y) / Y
F
y otherwise.
Observe that

F(x, ) has small bres for each x C
d
. By Corollary 3.1 there is > 0 such that for all
> 0 there is some a C
1
/2
such that the Lebesgue measure of
B(a) =
_
x C
d
:

FF

F(x, )
1
a
_

(0)
(C(x, )) (0)
_
is at least 1 .
Since each of the relations Y
F
is Borel, it is possible to nd a nite family of functions ( from C
d
to
C such that

FF
_
xC
d

_
Y
F
(x, ) F(x, )
1
G(x)
GG
_
dx <

d+1
2
d+1
.
It is then possible to nd a closed set D C
d
such that
(D) > (i + 1) + (i + 1) for each i < d (using Hypothesis 3.9)
(D) > 1
d
/2
d
(Y
F
(x, ) F(x, )
1
G(x)
GG
) < /2 for each x D
each G ( is continuous on D.
8
0
9


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
7
-
3
1







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


8 S. SHELAH AND J. STEPR

ANS
Condition 3 guarantees that

+
,
+(D) holds. It is therefore possible to apply the induction hypothesis
to
+
,
+
, ( and D to get a
0
C
1
/2
and : N R
+
such that

,
+
_
D

GG
G
1
a
0
_
.
Let 0 <

< (i) for i < d. Now choose a


1
C
1
/2
such that (B(a
1
)) > 1

. Then

,
+
_
D B(a
1
)

GG
G
1
a
0
_
holds by Lemma 3.6 and Condition 3 and note that by Condition 3

/2
_

FF

F(x, )
1
a
1
Y
F
(x, ) F(x, )
1
G(x)
GG
_
(0)
for x D B(a
1
). Let be dened by
(i) =
_

_
(i 1)

if d i > 0
(i 1) if d < i
/2 if i = 0
and dene Z to be the set of all (x, w) C
d+1
such that the following three conditions are satised:
(3.5) x D B(a
1
)

GG
G
1
a
0
(3.6) (F T)

F(x, w) a
1
(3.7) (F T) F(x, w) Y
F
(x, ) and F(x, w) ,= G(x)
It follows from Lemma 3.5 that
,
(Z). Hence it suces to observe that if (x, w) Z then F(x, w)
a
0
a
1
. To see this note that if F(x, w) =

F(x, w) this is immediate from 3.6. Otherwise F(x, w) Y
F
and hence it follows from 3.7 that there is some G ( such that F(x, w) = G(x). From 3.5 it can be
concluded that G(x) a
0
.
Notation 3.1. For the rest of the paper, x a pair of functions : N R
+
and : N R
+
such that
(3.8) (n) > (n + 1) and lim
n
(n) = 0
(3.9) (j) (j) + (j) < 1
(3.10) (j) (j) + (j + 1) < 1
(3.11) the range of is a dense subset of (0, 1).
Denition 3.3. For any d N and for X C
d
dene (X) =
,
(X) and dene

(X) to hold if
and only if there is some : N R
+
such that
,
(X) holds.
8
0
9


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
7
-
3
1







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


UNIFORMITY INVARIANTS 9
4. A preliminary norm
This section will introduce three norms
2
on subsets of C

these will be denoted ,

and . The
norms of [4] typically enjoy some form of sub-additivity but this will not be the case for any of these
three, at least not explicitly. Nevertheless, Lemma 4.2 can be considered a substitute for this. The
only norm used in the denition of the partial order in 6 will be . The role of the norm will be to
establish a connection between

and the results of 3. The norm

is an intermediary between
and and, furthermore, it has the advantage of allowing the compactness argument of 5 to work.
Notation 4.1. For any Polish space X let /(X) denote the spaces of compact subsets of X with the
Hausdor metric and let ((X) denote the space of continuous C-valued functions with the uniform
metric.
Denition 4.1. Dene to be a function from T(C

) to N by rst dening

(A) to be
_
d > 1 : (C /(C
d
))(F ((C))(x A)(C)

(F
1
x)
_
and let (A) = sup(

(A)). If

(A) = but A ,= then dene (A) = 0.


The following is a direct Corollary of Theorem 3.1.
Corollary 4.1. If > 0 then (C
1

) = .
Lemma 4.1. If d 1 then C /(C
d
) : (C) is a Borel set.
Proof. It suces to show (C) fails if and only if there is some nite subset ( Q
d
such that
C [(] =
_
(t
1
,t
2
,...,t
d
)C
[t
1
] [t
2
] . . . [t
d
]
and ([(]) also fails. To prove this proceed by induction on d. The case d = 1 is easy. Assuming the
result for d let C C
d+1
be compact and suppose that (C) fails. Using Denition 3.2 there is (
1
C

for some < (d) such that z C :


,
(C(x, )) [(
1
]. For x C [(
1
] it is possible to use the
induction hypothesis to nd a nite (
x
such that C(x, ) [(
x
] and
,
((
x
) fails. . The compactness
of C yields an integer k
x
such that C(y, ) [(
x
] for each y C such that y k
x
= x k
x
. Let U
x
denote this neighbourhood and choose a nite A C such that [(
1
] and U
a

aA
cover C. It is then
easy to extract the desired nite set from the cover [(
1
] C
d
U
a
[(
a
]
aA
.
Corollary 4.2. For any d 1 the set C /(C
d
) :

(C) is Borel.
Proof. According to Denition 3.1,

(C) holds if and only if there is some : N R


+
such that

,
(C) holds. But notice that can be assumed to be a constant function with rational value.
Notation 4.2. The notation z, F) will be introduced for any z C and any function from F : W
T(Q) to denote w W : z / [F(w)].
Corollary 4.3. Suppose that A C

and B : A T(C

). Then z C : (z, B) < j is analytic


for any integer j.
Proof. This is immediate from Denition 4.1, Lemma 4.1 and Corollary 4.2.
Lemma 4.2. Suppose that A C

and (A) = j 2. Then

(j2)
(z C : (z, F)) < j 1) < (j 1)
for each function F from A to C

(j1)/2
.
2
This is the terminology of [4]. The exact meaning of the term norm is not required for the arguments used here.
8
0
9


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
7
-
3
1







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


10 S. SHELAH AND J. STEPR

ANS
Proof. Let S = z C : (z, F)) < j 1 and assume that the lemma fails. In other words,
(j2)
(S)
(j 1). Let
E C /(C
j1
) ((C
j1
)
be the set of all triples (z, C, f) such that
(4.1) (C)
(4.2) (x z, F))

(C f
1
x)
It follows from Denition 4.1 that S is contained in the domain of E. From Lemma 4.1 and Corollary 4.2
it follows that E is a Borel set. From Corollary 4.3 it follows that S is measurable and so it is possible
to use the von Neumann Selection Theorem and Egeros Theorem to nd a closed set S

in the domain
of E and a continuous function T E with domain S

such that (S S

) < (j 2) (j 1). Let


T(s) = (C
s
, f
s
) and observe that the continuity of T guarantees that C

sS

s C
s
is a closed
set. Moreover,
(j1)
(S

)
(j2)
(S) (j 1). A calculation using Denition 3.2 reveals that
(C

) holds. Let g be dened on C

by g(c
1
, c
2
, . . . , c
j
) = f
c
1
(c
2
, c
3
, . . . , c
j
) recalling that j 2.
Since (A) j it is possible to nd x A such that

(g
1
x). Let
W = z S

(g
1
x(z, )) = z S

(g(z, )
1
x)
and note that it follows that
r

(W) (j 1). Since F(x) C


(j1)/2
it is possible to choose
w W F(x). Then x w, F) and so, by Condition 4.2 in the denition of E, it follows that

(g(w, )
1
x) fails since g(w, ) = f
w
. This contradicts that w W.
Denition 4.2. The norms and

will be dened for the subsets of C

by rst using induction to


dene an associated sequence of sets:
^
0
= ^

0
= A C

: A ,=
^
1
= ^

1
= A C

: [A] = C
(4.3) ^
j+1
= A C

: (F : A C
(j)/2
)
(j1)
(z C : z, F) / ^
j
)) < (j)
(4.4) ^

j+1
= A C

: (F : A C
(j)/2
)
(j1)
_
z C : z, F) / ^

j
)
_
< (j).
Then dene (A) to be the supremum of all j such that A ^
j
and

(A) to be the supremum of all


j such that A ^

j
.
It must be noted that ^
j
^
j+1
and ^

j
^

j+1
for each integer j and hence the supremum in
Denition 4.2 is taken over an initial segment of the integers. This will be used implicitly in what
follows.
Corollary 4.4. If A C

then (A)

(A).
Proof. Proceed by induction on j = (A). The case j = 0 is trivial but the case j = 1 is less so.
To see that (A) = 1 implies that A covers C let z C. Note that (C) holds by Condition 3.10
of Notation 3.1. Letting F be the function on C with constant value z it follows from the denition
of (A) = 1 that there is some x A such that

(F
1
x) holds. In particular, F
1
x ,= and so if
w F
1
x then z = F(w) x.
Therefore it can be assumed that 2 j +1 = (A) and that the lemma holds for j. In order to show
that

(A) j + 1 let F : A C

(j)/2
. By Lemma 4.2 it must be that

(j1)
(z C : (z, F)) < j) < (j).
From the induction hypothesis it follows that if

(z, F)) < j then (z, F)) < j. Hence

(j1)
(z C :

(z, F)) < j) < (j)


8
0
9


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
7
-
3
1







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


UNIFORMITY INVARIANTS 11
and this establishes that

(A) j + 1.
5. Finding finite sets with large norm
It will be shown that for any j there is a nite A such that

(A) j. This will establish that the


C(k) required in the denition of the partial order P in 6 actually do exist. Each of the next lemmas
is a step towards this goal.
Lemma 5.1. If A C

then

(A) (A).
Proof. This follows from the denitions by an argument using induction on j to show that if

(A)
j + 1 then (A) j + 1.
Lemma 5.2. If /
n

n=0
is an increasing sequence of nite subsets of C

then

_
n=0
/
n
_
= lim
n

(/
n
).
Proof. Proceed by induction on j to show that if

(/
n
) < j for each n then

n=0
/
n
) < j. For
j = 0 this trivial and if j = 1 this is simply a restatement of the compactness of C. Therefore assume
that j 1, that the lemma is true for j, that

(/
n
) j for each n yet

n=0
/
n
) > j. Let
F
n
: /
n
C

(j)/2
witness that

(/
n
) , j + 1. In other words, using Equality 4.4 of Denition 4.2,

(j1)
(S
n
) (j) where S
n
= z C :

(z, F
n
)) < j.
Claim 1. If
n

n=0
is a sequence of positive reals and and A
n

n=0
is a sequence of elements of C

1
then there are two increasing sequences of integers k
n

n=0
and m
n

n=0
such that, letting
D
n
= A
mn
Q[< k
n
]
the following hold:
(5.1) if i n then D
n
= D
i
Q[< k
n
]
(5.2)
_

_
i=n
[A
m
i
D
n
]
_
<
n
.
Moreover, the increasing sequence k
n

n=0
can be chosen from any given innite set K.
Proof. Let K be given and, using the fact that 0 < r < 1, let k
n

n=0
K be such that for each n

i=n
2
k
i
(1r)

n
/2
and then choose the sequence m
n

n=0
such that Conclusion 5.1 holds. Then Conclusion 5.2 follows
from the following inequalities

_
i=n
[A
m
i
D
n
]
_

_

_
i=n
[A
m
i
D
i
] [D
i+1
D
i
]
_

i=n
([A
m
i
D
i
]) + ([D
i+1
D
i
])

i=n
_
_

tAm
i
\D
i
2
|t|
+

tD
i+1
\D
i
2
|t|
_
_

i=n
_
_

tAm
i
\D
i
2
|t|r
2
|t|(1r)
+

tD
i+1
\D
i
2
|t|r
2
|t|(1r)
_
_
8
0
9


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
7
-
3
1







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


12 S. SHELAH AND J. STEPR

ANS

i=n
_
_
2
k
i
(1r)
_
_

tAm
i
\D
i
2
|t|r
+

tD
i+1
\D
i
2
|t|r
_
_
_
_

i=n
2
k
i
(1r)
(1 + 1)
n
.

Before continuing, let a


i

i=0
enumerate

n=0
/
n
and, without loss of generality, assume that /
n
=
a

n
=0
. Using the claim and its nal clause and a diagonalization argument, nd two increasing
sequences of integers k
n

n=0
and m
n

n=0
such that for each i and each n i, letting F
n,i
= F
mn
(a
i
)
Q[< k
n
], Conclusion 5.1 and Conclusion 5.2 of the claim hold for a summable sequence of . To be
more precise, F
n+1,i
is an end extension of F
n,i
and

_
j=n
_
F
m
j
(a
i
) F
n,i

_
< 2
in
for any n i. Dene F(a
i
) =

n=i
F
n,i
and note that F(a
i
) C

(j)/2
since F
n,i
C
(j)/2
for each n
and the F
n,i
are increasing with respect to n.
Now let S = z C :

(z, F)) < j. Because

_

_
n=0
/
n
_
j + 1
it follows that
(j1)
(S) < (j) and, hence, it is possible to choose Y such that
r

(S Y ) < (j) and


(Y ) < (j 1). Choose M so large that letting
W =

_
i=0
_
_

_
n=max(M,i)
_
F
mn
(a
i
) F
max(M,i),i

_
_
it follows that (W) < (j 1) (Y ).
Now dene F

(a
i
) by
F

(a
i
) =
_
F
M,i
if i M
F
i,i
if i > M
and note that F

(a
i
) F(a
i
). Now let S

m
= z C :

(z, F

/
m
)) < j and note that S

m
S

m+1
and they are all compact. Indeed,
S

m
=
_
BAm,

(B)<j
_
_

aAm\B
[F

(a)]
_
_
and each [F

(a)] is compact.
Claim 2. If z / W and i M then z, F

/
m
i
) z, F
m
i
).
Proof. Let a

/ z, F
m
i
) and assume that m
i
since otherwise it is immediate that a

/ z, F

/
m
i
).
There are two cases to consider. First assume that M. Then z [F
m
i
(a

)] but since z / W
and i M it follows that z / [F
m
i
(a

) F
M,
]. Hence z [F
M,
] = [F

(a

)]. In other words,


a

/ z, F

/
m
i
). If > M a similar argument works.
It follows from the claim that if z / W and i M then

(z, F

/
m
i
))

(z, F
m
i
)). Hence, if
i M and z S
m
i
W then

(z, F

/
m
i
)) < j. In other words, S
m
i
W S

m
i
W. Therefore,
8
0
9


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
7
-
3
1







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


UNIFORMITY INVARIANTS 13

(j1)(W)
(S

m
i
W)
(j1)(W)
(S
m
i
W) (j). Since each of the S

m
i
W are compact it
follows that

(j1)(W)
_

i=M
S

m
i
W
_
(j).
Let S

i=M
S

i
. Since (Y ) < (j 1) (W) it follows that
r

(S

(W Y )) (j).
Since
r

(S Y ) < (j) it is possible to select z S

(S W Y ). Then

(z, F

/
m
i
)) < j for
each i M and so the induction hypothesis guarantees that

_

_
i=M
z, F

/
m
i
)
_
< j.
But

_
i=M
z, F

/
m
i
) = z,

_
i=M
F

/
m
i
) = z, F

)
and so

(z, F

)) < j. Since F

(a) F(a) for each a it is immediate that z, F

) z, F) and so

(z, F)) < j. This contradicts that z / S.


Corollary 5.1. For any j N and > 0 there is a nite set A C
1

such that (A) j.


Proof. Combine Corollary 4.1 and Corollary 4.4 to conclude that

(C
1

) = . Then use Lemma 5.2


to nd a nite A C
1

such that

(A) j. Finally apply Lemma 5.1.


6. The forcing partial order
Using Corollary 5.1 let C(n) be a nite subset of C
1
2
n
such that (C(n)) n for each n N. Recalling
the notation concerning trees in 2, let P consist of all trees T such that:
(6.1) (t T)(i < [t[) t(i) C(i)
(6.2) (k )(t T)(s T) t s and (o
T
(s)) > k
and let this be ordered under inclusion. The methods of [4] can be used to establish that P is an

-bounding proper partial order.


Lemma 6.1. Let V W be models of set theory and suppose that G PV is generic over W. Then
W[G] [= (W [0, 1]) = 0.
Proof. If G P V is generic over W then let B
G

n=0
C(n) be the generic branch determined by
G. Note that ([B
G
(j)]) < 2
j
for each j and so (

j=0
[B
G
(j)]) < . Also, for every T P there is
some t T such that V [= (o
T
(t)) 1. Since (o
T
(t)) 1 is equivalent to o
T
(t) = C, and this is
absolute, it follows from genericity that C W

j=m
[B
G
(j)] for every m.
Denition 6.1. Let P

be the countable support product of copies of P.


Corollary 6.1. If
1
and G P

is a lter generic over V then in V [G] every set of reals of size


less than has Lebesgue measure 0.
In light of Corollary 6.1 the goal now is to establish that if P

is the countable support product of


copies of P and G P

is a lter generic over V then


r

(C V ) = 1 in V [G].
Lemma 6.2. Suppose that p P

and
p
P
C C


and < 1. Then there are A C and B C such that
r

(A) + (B) < 1 and q p such that


q ,
P
z [C] for each z C (A B).
8
0
9


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
7
-
3
1







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


14 S. SHELAH AND J. STEPR

ANS
Proof. Let 1 >
1
> and choose a monotonically increasing function
2
: N R such that
1
+
lim
n

2
(n) < 1. For later use, dene k(i) =

i
j=0
[C(j)[. The proof is based on a standard fusion
argument. Induction will be used to construct p
i
, S
i
,
i
, a
i
, b
i
and L
i
satisfying the following conditions:
(1) p
i
P

and p
0
p
(2) p
i+1
p
i
(3) S
i
: N is 0 at all but nitely many ordinals these will be denoted by D(S
i
)
(4) S
i
() = S
i+1
() for all but one
(5) S
i+1
() S
i
() + 1 for all
(6) L
i
N and L
i
< L
i+1
(7) p
i+1
()[L
i
] = p
i
()[L
i
] for each D(S
i
)
(8) if S
i
() > 0 there is a maximal antichain (
i,
p
i
()[< L
i
] such that (o
p
i
(
j
)
(s)) S
i
() for
each s (
i,
(9) for each

D(S
i
)
p
i
()[L
i
] there is some nite C
i,
and k
i,
N such that
p
i
)
P


C
i,
= C Q[< k
i,
] and C

C
i,
C

i

where p
i
) is dened by p
i
)() = p
i
()())
(10) a
i
C

i
for some
i
<
1
and a
i
a
i+1
(11) ([b
i
]) <
2
(i) and b
i
b
i+1
(12) for each z C [a
i
b
i
] there is T
,z,i
P for each D(S
i
) such that
(a) T
,z,i
p
i
()[ L
i
]
(b) if s T
,z,i
[< L
i
] (
j,
then (o
T
,z,i
(s)) S
j
()
(c) z / [C
i,
] for any

D(S
i
)
T
,z,i
[L
i
]
(d) T
,z,i
[L
i
] = T
,z,i+1
[L
i
]
(13) 2
i
k(L
i
)
2i
<
1

i
(14) lim
n
S
n
() = for each element of the domain of some p
i
Assuming this can be done, let A =

i=0
[a
i
] and B =

i=0
[b
i
] and dene
q() =

_
i=0
p
i
()[L
i
]
for each which is in the domain of some p
i
. It follows from Conditions 8 and 7 that q P

. Then
if z C [A B] let
q
z
() =

_
j=0
T
,z,j
and note that it follows from Condition 12b that q
z
P

. From Conditions 9 and 12c it follows that


q
z

P
z / C.
To see that the induction can be carried out, let i be given and suppose that p
i
, S
i
,
i
, a
i
, b
i
and L
i
satisfying the induction hypothesis have been chosen. Choose according to some scheme which will
satisfy Condition 14 and dene
S
i+1
() =
_
S
i
() if ,=
S
i
() + 1 if = .
Let p
0
p
i
be such that p
0
= p
i
if S
i
( ) > 0 and, otherwise, p
0
() = p
i
() for ,= and [ p
0
( )[L
i
][ = 1.
Let m S
i+1
( ) be such that
(6.3)
(m)
2
>
i
k(L
i
)
i
8
0
9


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
7
-
3
1







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


UNIFORMITY INVARIANTS 15
(6.4) (m)k(L
i
) <
1

i
(6.5) (m1) <

2
(i + 1)
2
(i)
k(L
i
)
and for each t p
0
( )[L
i
] there is some t

p
0
( ) such that t t

and such that


(6.6) (o
p
0
( )
(t

)) m+ 1.
Let L
i+1
> L
i
be so large that [t

[ < L
i+1
for each t p
0
( )[L
i
]. Then let p
1
p
0
be such that
(
i+1,
= t

: t p
0
( )[L
i
] is a maximal antichain in p
1
( )
if t p
1
( ) and L
i
[t[ L
i+1
then either
[o
p
1
( )
(t)[ = 1 or
o
p
1
( )
(t) = o
p
0
( )
(t)
p
1
() = p
0
() for ,= .
Let (
i+1,
= (
i,
if ,= . Let p
2
p
1
be such that p
2
() = p
1
() for / D(S
i
) and if
D(S
i
) and t p
2
() and L
i
[t[ L
i+1
then [o
p
2
()
(t)[ = 1. Let
i+1
> 0 be so small that
2
i+1
k(L
i+1
)
2(i+1)
<
1

i

i
k(L
i
).
Choose p
i+1
p
2
such that p
i+1
()[L
i+1
] = p
2
()[L
i+1
] for each D(S
i+1
) this implies that
Condition 7 holds and for each

D(S
i+1
)
p
i+1
()[L
i+1
] there is some C
i+1,
and k
i+1,
such that
p
i+1
)
P


C
i+1,
= C Q[k
i+1,
] and C

C
i+1,
C

i+1
.
For each t p
i
( )[L
i
] and x o
p
i
( )
(t

) and

D(S
i
)\{ }
p
i
()[L
i
] let (t, x, ) be the unique element
of

D(S
i+1
)
p
i+1
()[L
i+1
] such that (t, x, ) () for each D(S
i
) and (t, x, ) t

and
(t, x, )([t

[) = x. Let

(t, x, ) be the unique element of

D(S
i
)
p
i
()[L
i
] such that

(t, x, ) = if
S
i
( ) = 0 and

(t, x, ) and

(t, x, )( ) = t if S
i
( ) > 0.
Note that C
i+1,(t,x,)
C
i,

(t,x,)
C

i
by Condition 9. Dene F
t
: o
p
i+1
( )
(t

) C

i
k(L
i
)
i by
F
t
(x) =
_

Q
D(S
i
)\{ }
p
i
()[L
i
]
C
i+1,(t,x,)
C
i,

(t,x,)
and observe that F
t
(x) C
(m)/2
by 6.3 since

D(S
i
)
p
i
()[L
i
]

k(L
i
)
|D(S
i
)|
k(L
i
)
i
.
By 6.6 it follows that

(m1)
(z C : (z, F
t
)) < m) < (m)
for each t p
i
( )[L
i
]. Let b
t
i+1
be such that (b
t
i+1
) < (m 1) and such that there is a
t
i+1
C
(m)
such that
(6.7) [a
t
i+1
] z C : (z, F
t
)) < m [b
t
i+1
].
It follows from 6.5 and Induction Hypothesis 11 that if b
i+1
is dened to be b
i

tp
i
( )[L
i
]
b
t
i+1
then
Condition 11 is satised by b
i+1
. Similarly, if a
i+1
is dened to be a
i

tp
i
( )[L
i
]
a
t
i+1
then a
i+1
C

i+1
where
i+1
=
i
+ (m)k(L
i
) <
1
by 6.4 and Induction Hypothesis 10.
In order to verify that Condition 12 holds let z C [a
i+1
b
i+1
]. Let T
z,,i+1
= T
z,,i
if ,= and
let T
z, ,i+1
be the set of all s p
i+1
( )[ L
i+1
] such that if t

s then s([t

[) z, F
t
). In order to
show that Condition 12b holds it suces to show that
(o
T
,z,i+1
(t

)) m > S
i+1
( )
8
0
9


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
7
-
3
1







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


16 S. SHELAH AND J. STEPR

ANS
for any t p
i
( )[L
i
]. For any given t this follows from the fact that z / [a
t
i+1
b
t
i+1
] and 6.7.
In order to show that Condition 12c holds let

D(S
i+1
)
T
,z,i
[L
i+1
]. Let = (D(S
i
) ).
Let t p
i
( )[L
i
] be such that t ( ) and note that the denition of T
,z,i
guarantees that ( )([t

[)
z, F
t
). In other words, z / [F
t
(( )([t

[))] [C
i+1,(t,( )(|t

|), )
C
i,

(t,( )(|t

|), )
]. Moreover, since

(t, ( )([t

[), )

D(S
i
)
p
i
()[L
i
] it follows that z / [C
i,

(t,( )(|t

|), )
] by the Induction Hypothe-
sis 12c. Hence z / [C
i+1,
].
Corollary 6.2. If G P is generic over V then V [G] [=
r

(V C) = 1.
Proof. Suppose not and that
p
P
C C

and V C [C]
and < 1. Using Lemma 6.2 nd A C and B C such that
r

(A) + (B) < 1 and q p such


that q ,
P
z [C] for each z C (A B). Choose any z C (A B) and q

q such that
q

P
z / [C].
7. Remarks and open questions
One might expect that the methods developed here could be used to prove that for any two reals s
and t such that 0 < t < s < 1 it is consistent that sets of size
1
are null with respect to s-dimensional
Hausdor measure but that this is not so for t-dimensional Hausdor measure. While this is true
of most of the argument there are some slippery spots. For example. the use of Y
F
in the proof of
Lemma 3.4 assumes the -niteness of Lebesgue measure. Lemma 3.2 might also pose some challenges
to generalization. Let n
s
denote the least cardinal of a set which is not null with respect to s-dimensional
Hausdor measure Hence the following question remains open:
Question 7.1. Is it consistent that 0 < t < s < 1 and n
t
< n
s
?
Question 7.2. Is it consistent that 0 < u < v < w < 1 and n
u
< n
v
< n
w
?
Question 7.3. How big can the cardinality of n
s

s(0,1)
be?
However, the main open problem in this area still remains Question 1.1. It would be interesting to
know what the answer to this question is in the model described in 6.
References
[1] David Fremlin. Problem sheet. http://www.essex.ac.uk/maths/staff/fremlin/problems.htm, 2002.
[2] Martin Goldstern and Saharon Shelah. Many simple cardinal invariants. Archive for Mathematical Logic, 32:203221,
1993.
[3] Pertti Mattila. Geometry of sets and measures in Euclidean spaces, volume 44 of Cambridge Studies in Advanced
Mathematics. Cambridge University Press, Cambridge, 1995.
[4] Andrzej Roslanowski and Saharon Shelah. Norms on possibilities. I. Forcing with trees and creatures. Mem. Amer.
Math. Soc.
[5] T. Salisbury and J. Steprans. Hausdor measure and Lebesgue measure. Real Anal. Exchange, 22:265278, 1996/97.
[6] J. Steprans. Cardinal invariants associated with Hausdor capacities. In Proceeding of the BEST Conferences 1-3,
volume 192 of Contemporary Mathematics, pages 157184. AMS, 1995.
[7] J. Steprans. Unions of rectiable curves and the dimension of banach spaces. J. Symbolic Logic, 64(2):701726, June
1999.
8
0
9


r
e
v
i
s
i
o
n
:
2
0
0
3
-
0
7
-
3
1







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


UNIFORMITY INVARIANTS 17
Department of Mathematics, Rutgers University, Hill Center, Piscataway, New Jersey, U.S.A.
08854-8019
Current address: Institute of Mathematics, Hebrew University, Givat Ram, Jerusalem 91904, Israel
E-mail address: shelah@math.rutgers.edu
Department of Mathematics, York University, 4700 Keele Street, Toronto, Ontario, Canada M3J
1P3
E-mail address: steprans@yorku.ca