In this lecture we verify that the analytic sets are Lebesgue measurable (LM) and have the Baire property (BP). Since both properties are closed under complements, they also hold for the class of co-analytic setsΠΠ11.
The analytic sets also have the perfect subset property (PS).
For Borel sets, one proves (LM) and (BP) by showing that the class of sets having (LM) (or (BP), respectively) forms a σ-algebra and contains the open sets. For the analytic sets, this method is no longer available. We can, however, prove a similar property with respect to the Souslin operation A, which can be seen as an extension of basic set theoretic operations into the uncountable.
More specifically, we will show the following.
The Souslin operation A is idempotent, i.e. AAΓ=AΓ. This implies that the analytic sets are closed under A.
The family of sets with (LM) (or (BP), respectively), is closed under the Souslin operation. Since the closed sets have both properties, and the Souslin operator is clearly monotone on classes, this yields the desired regularity results.
Now we contract the two function quantifiers to a single one, using a (computable) homeomorphism NN×NN, and the two universal number quantifiers into a single one using the paring function ⟨.,.⟩. Then A can be characterized as
z∈A⟺∃γ∀k(z∈Rγ↾k)
where Rσ=Qφ(σ),ψ(σ)∈Γ for suitable coding functions φ,ψ. We leave an explicit definition of these coding functions as an exercise.
We start with a lemma that essentially says that we can envelop any
set with a smallest (up to measure 0) measurable set.
Proof
Suppose first that λ∗(A)<∞. For every n≥0, there exists an open set On⊇A with λ∗(On)=λ(On)<λ∗(A)+1/n. Then B=⋂nOn is measurable, and λ(B)=λ∗(A). Furthermore, if A⊆B′⊆B, then λ∗(A)≤λ∗(B′)≤λ∗(B).
If λ∗(A)=∞, let An=A∩[m,m+1) for m∈Z. Then λ∗(Am)≤1, and we can choose Bm⊇Am measurable such that λ∗(Bm)=λ∗(Am). Then B=⋃m∈ZBm has the desired property.
We now apply the lemma to show that Lebesgue measurability is closed
under the Souslin operation. The basic idea is to approximate the local
“branches” of the Souslin operation on a Souslin scheme by measurable
sets from outside, in the sense of the lemma.
It turns out that the total error we make by this approximation is
negligible, and hence the overall result of the Souslin operation
differs from a measurable set only by a nullset and hence is
measurable.
Proof
Let A=(Aσ) be a Souslin scheme with each Aσ measurable. We can assume that (Aσ) is regular. For each σ∈N<N we let
Aσ=α⊃σ⋃n∈N⋂Aα↾n⊆Aσ.
Note that A∅=AA.
By the previous lemma, there exist measurable sets Bσ⊇Aσ so that for every measurable B⊇Aσ, Bσ∖B is null.
By replacing Bσ with Bσ∩Aσ, we can further assume Bσ⊆Aσ, and also that (Bσ) is a regular Souslin scheme.
Now let Cσ=Bσ∖⋃n∈NBσ⌢⟨n⟩. Each Cσ is a nullset, by the choice of the Bσ and the fact that Aσ=⋃n∈NAσ⌢⟨n⟩⊆⋃n∈NBσ⌢⟨n⟩ . Hence C=⋃σCσ is a nullset, too.
It remains to show that
B∅∖C⊆A∅=AA,
for this implies B∅∖A∅⊆C is null, which in turn implies that A∅ is Lebesgue measurable (since it differs from a measurable set by a nullset).
So let x∈B∅∖C. Since x∈C∅, there is an α(0) with x∈B⟨α(0)⟩.
Given α↾n with x∈Bα↾n, we can choose α(n) so that x∈Bα↾n+1. This is possible because x∈Cα↾n. This way we construct α∈NN with
x∈n⋂Bα↾n⊆n⋂Aα↾n⊆A∅.
Proof
We have, as the Souslin operator is monotone on classes,
The previous proof is general enough to work for other kinds of
measures on arbitrary Polish spaces.
Given a Polish space X, a Borel measure on X is a
countably additive set function μ defined on a σ-algebra of the
Borel sets in X. A set is μ-measurable if it can be
represented as a union of a Borel set and a μ-nullset. A measure μ
is σ-finite if X=⋃nXn, where Xn is
μ-measurable with μ(Xn)<∞. Lebesgue
measure is σ-finite Borel measure on the Polish space R.
A set A⊆X is universally measurable if it is
μ-measurable for every σ-finite Borel measure on X.
Inspecting the proof of Proposition 1, we
see that it works for the Baire property as well (with “measure
0” replaced by “meager”, of course), provided we can
prove a Baire category version of Lemma 1.
Proof
Let U1,U2,… be an enumeration of countable base of the
topology for X.
Given A⊆R set
A∗:={x∈R:∀i(x∈Ui⇒Ui∩A not meager)}.
Note that A∗ is closed: If x∈A∗, then there exists i with x∈Ui&Ui∩A null. If y∈Ui, then y∈A∗, since Ui∩A is null.
Hence Ui⊆¬A∗.
We have
A∖A∗=⋃{A∩Ui:A∩Ui meager},
which is a countable union of meager sets and hence meager.
If we let B=A∪A∗=A∗∪(A∖A∗), then B is a
union of a meager set and a closed set and hence has the Baire
property.
Now assume B′⊇A has the Baire property. Then C=B∖B′ has the Baire property, too. Suppose C is not meager,
then Ui∖C is meager for some i, and hence also Ui∩A⊆(Ui∖C). Besides, Ui∩C=∅, for otherwise Ui⊆Ui∖C would be
meager. Thus there exists x∈Ui with x∈A∗, which by
definition of A∗ implies that Ui∩A is not meager, a contradiction.
By adapting the proof of Proposition 1, we
obtain the Baire category version and hence can deduce that analytic sets
have the Baire property.