156 (1998)
Strongly meager sets and subsets of the plane
by
Janusz P a w l i k o w s k i (Wrocław)
Abstract. Let X ⊆ 2
ω. Consider the class of all Borel F ⊆ X × 2
ωwith null vertical sections F
x, x ∈ X. We show that if for all such F and all null Z ⊆ X, S
x∈Z
F
xis null, then for all such F , S
x∈X
F
x6= 2
ω. The theorem generalizes the fact that every Sierpi´ nski set is strongly meager and was announced in [P].
A Sierpiński set is an uncountable subset of 2
ωwhich meets every null (i.e., measure zero) set in a countable set. Such sets may not exist, but they do, e.g., under the Continuum Hypothesis.
A strongly meager set is a subset of 2
ωwhose complex algebraic sum with null sets cannot give 2
ω.
Answering a question of Galvin I proved in [P] that every Sierpi´ nski set is strongly meager (see [M] and [P] for more about Galvin’s question).
Since Sierpiński sets may not exist, this result is somewhat defective. Here we prove its “absolute” version, which seems to be of independent inter- est. The paper is an elaboration of the Note given at the end of [P]. A different elaboration, using “small sets” of [B] and closely following my lecture at Cantor’s Set Theory meeting, Berlin 1993, is given in [BJ] in Section 5 (repeated in [BJ1]). There is, however, a major gap in the expo- sition in [BJ], namely, in the proof of Lemma 5.5, where one really needs a sort of Kunugui–Novikov theorem (see the proof of Lemma 4 below).
Also it seems reasonable to avoid “small sets” because they do not form an ideal.
Let X ⊆ 2
ω. Consider the class of all Borel F ⊆ X ×2
ωwith null vertical sections F
x, x ∈ X. If for all such F , S
x∈X
F
xis null, resp. 6= 2
ω, we say that X ∈ Add, resp. X ∈ Cov. (See [PR] for an explanation of this notation.) Here Borel means relatively Borel. It is useful to remember that for a Borel F ⊆ X × 2
ω, the function x 7→ µ(F
x) is Borel. In particular, any Borel
1991 Mathematics Subject Classification: 04A15, 03E15.
Partially supported by KBN grant 2 P03A 011 09.
[279]
subset of Y × 2
ω, Y ⊆ X, with all vertical sections null, extends to a Borel subset of X × 2
ωwith all vertical sections null.
Theorem. Suppose that every null subset of X ⊆ 2
ωis in Add. Then X ∈ Cov.
The theorem implies that every Sierpi´ nski set is strongly meager as fol- lows. Suppose X ⊆ 2
ωis a Sierpi´ nski set. Let D ⊆ 2
ωbe null. We want to see that D + X 6= 2
ω. Consider
F = [
x∈X
{x} × (D + x).
Clearly F is a Borel subset of X × 2
ωand all its vertical sections are null.
Also, every null subset of X, being countable, is in Add. So, by the theorem, S
x∈X
F
x6= 2
ω. But D + X = S
x∈X
F
x.
Notation. Given a set K and A ⊆ 2
K, let µ(A) be the measure of A in the product measure arising from assigning to each point in {0, 1} weight 1/2. Note that if K is finite, then µ(A) = |A| · 2
−|K|.
For K ⊆ L ⊆ ω and A ⊆ 2
Llet [A] = {t ∈ 2
ω: t|L ∈ A} and let A|K = {t|K : t ∈ A}. Likewise, for A ⊆ (2
L)
klet [A] = {ht
1, . . . , t
ki ∈ (2
ω)
k: ht
1|L, . . . , t
k|Li ∈ A}.
Clearly µ(A) = µ([A]). Note that any clopen subset of 2
ωcan be written as [A] for some A ⊆ 2
n. Also for A ⊆ 2
nand m > n, [A] = [B], where B = {τ ∈ 2
m: τ |n ∈ A}.
For σ ∈ ω
<ωof length n + 1 , let σ
∗be σ|n.
We use the following abbreviations:
∃
∞— there exist infinitely many,
∀
∞— for all but finitely many, W
n
— T
m
S
n>m
, V
n
— S
m
T
n>m
.
For F ⊆ X × T and x ∈ X let F
x= {t ∈ T : hx, ti ∈ F }. Likewise, for F ⊆ X × S × T , x ∈ X, s ∈ S, let F
xs= {t ∈ T : hx, s, ti ∈ F }, etc. In particular, if F ⊆ X × (2
ω)
ω, t
0, . . . , t
n∈ 2
ω, then
F
xt0...tn= {ht
n+1, . . .i ∈ (2
ω)
ω: hx, t
0, . . . , t
n, t
n+1, . . .i ∈ F }.
Let F [X] = S
x∈X
F
x. If F has all sections F
x, x ∈ X, null, we say that F is X-null.
The following simple lemma is crucial.
Lemma 1. Let every null subset of X be in Add. Suppose F ⊆ X × (2
ω)
ωis Borel and X-null. Then, given null Y ⊆ X, there exist t ∈ 2
ωand Borel
null Z ⊆ X \ Y such that F
xt, x 6∈ Z, are null.
P r o o f. Let
G = {hx, ti ∈ X × 2
ω: µ(F
xt) > 0}.
Then G is Borel and X-null. By Fubini’s theorem find t ∈ 2
ω\ G[Y ] such that
Z = {x : hx, ti ∈ G}
is null. (This is possible because G[Y ] is null.)
We shall need the following property B, which is a Borel version of prop- erty H of Hurewicz (for more see [FM], [PR]):
X ⊆ 2
ωhas property B if, given for each n ∈ ω a Borel cover {U
kn}
k∈ωof X, there exist k
n’s such that X ⊆ V
n
S
k≤kn
U
kn.
It is not hard to see that we can use increasing covers in this definition and write X ⊆ V
n
U
knn. Also, easily, X has property B iff for any Borel function f : X → ω
ω, f [X] is dominated. (Y ⊆ ω
ωis dominated if there exist z ∈ ω
ωsuch that ∀
∞n y(n) < z(n), for all y ∈ Y .) Moreover, it is enough to consider only f for which all f (x) are increasing.
The following lemmas are well known.
Lemma 2. If all null subsets of X ⊆ 2
ωhave property B, then X has property B.
P r o o f. Let {U
kn}
k∈ω, n ∈ ω, be increasing Borel covers of X. Find k
n’s with µ
∗(X \ U
knn) < 2
−n. Let Z = X \ V
n
U
knn. Then Z is null, so it has property B. Thus Z ⊆ V
n
U
lnnfor some l
n’s. It follows that X ⊆ V
n
U
max(kn n,ln).
Lemma 3. (1) If A ⊆ 2
ωis null then for any sequence {ε
n} of positive reals there exists an increasing sequence {a
n} ∈ ω
ωtogether with sets B
n⊆ 2
anof measure ≤ ε
n, n ∈ ω, such that A ⊆ W
n
[B
n].
(2) If a
n∈ ω and B
n⊆ 2
an, n ∈ ω, are such that P
n
µ(B
n) < ∞, then A = W
n
[B
n] is null. If moreover K ⊆ ω is such that P
n
µ(B
n)·2
|an∩K|< ∞, then also A|(ω \ K) is null (in 2
ω\K).
P r o o f. We prove the first part, the second is straightforward. Given null A ⊆ 2
ωand ε > 0, we can cover A by an open set of measure < ε/2
n, which next can be split into disjoint clopen sets. In this way we can find clopens C
i, i ∈ ω, such that A ⊆ W
i
C
iand P
i
µ(C
i) < ε.
Suppose now that {ε
n} is a sequence of positive reals. Use the above to find clopens C
i, i ∈ ω, such that A ⊆ W
i
C
iand P
i
µ(C
i) < ε
0. Next find an increasing sequence {i
n} such that P
i≥in
µ(C
i) < ε
n+1. Finally, let A
0= S
i<i0
C
iand for n > 0 let A
n= S
in−1≤i<in
C
i. Then µ(A
n) < ε
nand A ⊆ W
n
A
n. Each A
n, being clopen, is of the form [B
n] for some B
n⊆ 2
an.
We can easily arrange that a
n+1> a
n.
It follows from Lemma 3 that, given {b
n} ∈ ω
ωand null A ⊆ 2
ω, there is an increasing {a
n} ∈ ω
ωsuch that
∀
∞n |a
n∩ K| ≤ b
n⇒ A|(ω \ K) is null.
Lemma 4. Suppose that F ⊆ X × 2
ωis Borel X-null. Let {ε
n} be a sequence of positive reals. Then there exist for each n a countable Borel partition U
nof X together with integers a
nUand sets A
nU⊆ 2
anUof measure
≤ ε
n, U ∈ U
n, such that
F ⊆ _
n
[
U ∈Un
U × [A
nU].
If additionally X has property B, we can require that for some increasing {a
n} ∈ ω
ωall a
nU, U ∈ U
n, equal a
n.
P r o o f. This is a parametrized version of the first part of Lemma 3. We indicate the main steps.
For any ε > 0 there are Borel sets W
iand clopens C
i, i ∈ ω, such that F ⊆ W
i
W
i× C
iand for all x ∈ X, P
i∈K(x)
µ(C
i) < ε, where K(x) = {i : x ∈ W
i}.
This follows from the following facts:
• any Borel subset of X × 2
ωwith open vertical sections can be written as a union of countably many disjoint sets of the form W × C, W Borel, C clopen (a theorem of Kunugui and Novikov, see [K]);
• for any Borel B ⊆ X × 2
ωand ε > 0, there exists a Borel A ⊆ X × 2
ω, B ⊆ A, such that vertical sections of A are open and µ(A
x\ B
x) < ε (the sets with such a covering property form a monotone family that includes all finite unions of Borel rectangles).
Note that the function x 7→ P
i∈K(x)\j
µ(C
i) is Borel. (Because the function µ(C
i)1
Withat takes µ(C
i) on W
iand 0 outside is Borel and P
i∈K(x)\j
µ(C
i) = P
i≥j
µ(C
i)1
Wi(x).) It follows that for any δ > 0 we can find a countable Borel partition U of X and numbers j
U∈ ω, U ∈ U, such that on each U the mapping x 7→ K(x) ∩ j
Uis constant and P
i∈K(x)\jU
µ(C
i) < δ.
Using this find Borel sets U
σand integers j
σ, σ ∈ ω
<ω, such that
• U
∅= X, j
∅= 0,
• U
σis partitioned into U
σ_k’s, k ∈ ω,
• if |σ| > 0, then j
σ∗< j
σand x 7→ K(x) ∩ [j
σ∗, j
σ) is constant on U
σ,
• if |σ| > 0, then on U
σ, X
i∈K(x)∩[jσ∗,jσ)
µ(C
i) < ε
|σ|−1.
If |σ| > 0, let
B
σ= [
i∈K(x)∩[jσ∗,jσ)
C
i, x ∈ U
σ.
This is a clopen set of measure < ε
|σ|−1, so we can find a
σ∈ ω and A
σ⊆ 2
aσof measure < ε
|σ|−1such that B
σ= [A
σ].
Now just note that _
i
W
i× C
i⊆ _
n>0
[ {U
σ× [A
σ] : σ ∈ ω
n}.
Up to some enumeration, we are done.
The following lemma is a version of Miller’s [M1] result that additivity of measure is below number b. (See also [PR].)
Lemma 5. Add ⊆ B.
P r o o f. Let Y ∈ Add. Let Y 3 y 7→ y ∈ ω
ωbe Borel with all y’s increasing. Define F ⊆ Y × 2
ωby
t ∈ F
y⇔ ∃
∞n ∀i < n t(y(n) + i) = 0.
Then F is Borel and Y -null, so A = F [Y ] is null. Use Lemma 3 to find an increasing sequence {a
n} such that
∀n |K ∩ a
n| ≤ n(n − 1)/2 ⇒ A|(ω \ K) is null.
We claim that {a
n} dominates all y’s. Indeed, suppose that ∃
∞n y(n) ≥ a
n. Consider
K = [
{[y(n), y(n) + n) : y(n) ≥ a
n}.
Then ∀n |K ∩ a
n| ≤ n(n − 1)/2 (we take to K below a
nat most n − 1 intervals).
It follows that A|(ω \ K) is null. This is a contradiction because F
y⊆ A and F
y|(ω \K) is 2
ω\K. (Any element of 2
ω\Kcan be extended to an element of 2
ωwhich on infinitely many intervals [y(n), y(n)+n) is constantly zero.)
Proof of theorem. By Lemmas 2 and 5, X ∈ B. Let F ⊆ X × 2
ωbe Borel X-null. We seek a point outside F [X]. Let Q = {t ∈ 2
ω: ∀
∞n t(n) = 0}.
Enlarging F if necessary we can assume that for all x, F
x= F
x+ Q. Use Lemma 4 to find an increasing {a
n} ∈ ω
ωtogether with a sequence {U
n} of countable Borel partitions of X such that for some A
nU⊆ 2
an, U ∈ U
n, of measure ≤ 2
−n,
F ⊆ _
n
[
U ∈Un
U × [A
nU].
Let B
xnbe A
nUfor the unique U ∈ U
nthat covers x.
Say that σ
0, . . . , σ
k∈ 2
anhave a diagonal in A ⊆ 2
anif for some n
0≤ . . . ≤ n
k−1≤ n,
σ
0|a
n0∪ σ
1|[a
n0, a
n1) ∪ . . . ∪ σ
k|[a
nk−1, a
n) ∈ A.
Say that t
0, . . . , t
k∈ 2
ωhave a diagonal in A ⊆ 2
anif t
0|a
n, . . . , t
k|a
ndo.
Define E ⊆ X × (2
ω)
ωby
ht
0, t
1, . . .i ∈ E
x⇔ ∃k ∃
∞n t
0, . . . , t
khave a diagonal in B
xn. Claim 1. E is Borel and X-null.
P r o o f. Let
B
nx(k) = {hσ
0, . . . , σ
ki ∈ (2
an)
k+1: σ
0, . . . , σ
khave a diagonal in B
nx}.
Then |B
xn(k)| ≤ 2
an−n2
ank· (1 + n)
k. Indeed, there are ≤ (1 + n)
kpossi- ble sequences n
0, . . . , n
k−1, and for each sequence we have |B
xn| times 2
ankpossible choices for hσ
0, . . . , σ
ki.
So
µ([B
xn(k)]) = |B
xn(k)|/2
an(k+1)≤ 2
−n(1 + n)
k. It follows that
µ(E
x) ≤ X
k
Y
m
X
n≥m
2
−n(1 + n)
k= 0.
Claim 2. There exist {t
i} ⊆ 2
ωand a Borel partition {X
i} of X such that each
E
xti...tk, x ∈ X
i, k ≥ i, is null.
P r o o f. Apply Lemma 1 with Y = ∅ and F = E to find t
0∈ 2
ωand Borel null X
1⊆ X such that the following sets are null:
E
x, x ∈ X
1, E
xt0, x 6∈ X
1. Next apply Lemma 1 to Y = X
1and
F = [
x∈X1
{x} × E
x∪ [
x6∈X1
{x} × E
xt0to get t
1∈ 2
ωand Borel null X
2⊆ X \ X
1such that the following sets are null:
E
x, x ∈ X
2,
E
xt1, x ∈ X
1,
E
xt0t1, x 6∈ X
1∪ X
2.
Similarly find t
2∈ 2
ωand Borel null X
3⊆ X \ (X
1∪ X
2) such that the following sets are null:
E
x, x ∈ X
3, E
xt2, x ∈ X
2, E
xt1t2, x ∈ X
1,
E
xt0t1t2, x 6∈ X
1∪ X
2∪ X
3, etc. Finally, set X
0= X \ S
i>0
X
i.
It follows from Claim 2 that for x ∈ X
i, ht
i, t
i+1, . . .i 6∈ E
x.
Otherwise we would have E
xti...tk= (2
ω)
ωfor some k.
Thus for x ∈ X
iand k ≥ i,
∀
∞n t
i, . . . , t
khave no diagonal in B
xn. For all k and n let
V
nk= [
i>k
X
i∪ [
i≤k
{x ∈ X
i: ∀m ≥ n t
i, . . . , t
khave no diagonal in B
xm}.
Then for all k, V
nk’s form an increasing Borel cover of X. By X ∈ B, there is an increasing sequence {n
k} such that
X ⊆ ^
k
V
nkk.
Let
t = t
0|a
n0∪ t
1|[a
n0, a
n1) ∪ t
2|[a
n1, a
n2) ∪ . . . . Claim 3. t 6∈ F [X].
P r o o f. Fix x ∈ X
i. Since ∀
∞k x ∈ V
nkk, for all sufficiently large k ≥ i and n ≥ n
k, t
i, . . . , t
khave no diagonal in B
xn. Hence,
t
i|a
ni∪ t
i+1|[a
ni, a
ni+1) ∪ . . . ∪ t
k|[a
nk−1, a
n) 6∈ B
nx, and thus
t
i|a
ni∪ [
k>i
t
k|[a
nk−1, a
nk) 6∈ _
n
[B
xn].
It follows that t 6∈ F
x.
Note. We have really proved that if X ∈ B has all its null subsets in Cov,
then X ∈ Cov. The crucial Lemma 1 goes through because if Y ∈ Cov, then
for all Borel Y -null F ⊆ Y × 2
ω, µ
∗(2
ω\ F [Y ]) = 1. (Otherwise we could
find in F [Y ] a perfect set P of positive measure. Then D = F ∩ (Y × P )
would be a Borel Y -null subset of Y × P such that D[Y ] = P . This would
yield a similar subset of Y × 2
ω.)
Note also that if X ∈ B ∩ Cov, then 2
ω\ F [X] contains a perfect set for all X-null F ⊆ X × 2
ω. (It is enough to require in Lemma 4 that A
nU’s have measure ≤ 2
−2nand consider B
Un= {σ ∈ 2
an: ∃τ ∈ A
nUσ|K = τ |K}, where K is a fixed co-infinite subset of ω such that ∀n |a
n\ K| ≤ n. Then B
Unis a subset of 2
anof measure ≤ 2
−n. If X × {t} avoids W
n
S
U
U × [B
Un], then {s ∈ 2
ω: s|K = t|K} is a perfect set disjoint from F [X].)
We cannot drop B in the above remark. If we add ω
2random reals to a model of CH then the ground model reals constitute a counterexample. (Use the fact that a random real does not add a perfect set of random reals.)
We cannot require in the theorem that X ∈ Add. It is enough to take for X a Sierpi´ nski set and for F the diagonal in X ×X. There is however no ZFC example for this. Indeed, suppose the Dual Borel Conjecture holds, i.e. all strongly meager sets, hence also all Cov sets, are countable. Suppose also that every uncountable set has an uncountable null subset. (Both assumptions are true when ω
2Cohen reals are added to a model of CH, see [C].) If all null subsets of X are in Cov, then they are all countable by the first assumption.
So X has no uncountable null subsets, and thus X itself is countable by the second assumption. It follows that X ∈ Add.
Suppose that all null subsets of X are in Cov. Does it follow that X is strongly meager? We have the following partial result:
Proposition. Let X have property B. Let D = W
k
[B
k], B
k⊆ 2
Lk, where L
k⊆ ω, k ∈ ω, are pairwise disjoint. Suppose for every finite F ⊆ 2
ω, D + (X ∩ (D + F )) 6= 2
ω. Then D + X 6= 2
ω.
P r o o f. Choose t
0∈ 2
ωand inductively t
n∈ 2
ωso that t
n6∈ D + (X ∩ (D + {t
0, t
1, . . . , t
n−1})).
Then for all x ∈ X,
∀
∞n x 6∈ D + t
n.
Indeed, if x ∈ D + t
nand m > n, then t
m6∈ D + x, so x 6∈ D + t
m. Let
U
kn= {x : x 6∈ D + t
n⇒ ∀m ≥ k x|L
m6∈ B
m+ t
n|L
m}.
Then for all n, U
kn’s form an increasing Borel cover of X. So X ⊆ V
n
U
knnfor some increasing sequence {k
n}. Then for all x,
∀
∞n ∀k ≥ k
nx|L
k6∈ B
k+ t
n|L
k(remember that ∀
∞n x 6∈ D + t
n).
So for all x,
∀
∞n ∀k ≥ k
nt
n|L
k6∈ B
k+ x|L
k.
Let t ∈ 2
ωbe such that for all n, [
kn≤k<kn+1