Skip to article frontmatterSkip to article content

We will later see that the analytic sets correspond to the sets definable by means of Σ11\Sigma^1_1 formulas, that is formulas in the language of second order arithmetic that have one existential function quantifier.

Therefore, we will denote the analytic subsets of XX also by

Σ11(X).\PS{1}(X).

Here are some simple properties of analytic sets.

Proof

(i): This follows directly from Corollary 1 of the previous section.

(ii): The composition of continuous mappings is continuous.

(iii): Let AnA_n be analytic and fn:NNXf_n:\Baire \to X such that fn(NN)=Anf_n(\Baire) = A_n. Define f:NNXf: \Baire \to X by

f(m,α)=fn(α).f(m,\alpha) = f_n(\alpha).

Then ff is continuous and f(NN)=nAnf(\Baire) = \bigcup_n A_n.

We can use our previous results about Borel sets to give various equivalent characterizations of analytic sets.

Proof

(i) \Leftrightarrow (ii): Follows from the Lusin-Souslin Theorem and Proposition 1 (ii).

(ii) \Leftrightarrow (iii): Follows from the corollary to the Lusin-Souslin Theorem and Proposition 1 (ii).

(i) \Rightarrow (iv): Let f:NNXf:\Baire \to X be continuous, f(NN)=Af(\Baire) = A. Then

xA    α  (α,x)Graph(f),x \in A \iff \exists \alpha \; (\alpha,x) \in \Op{Graph}(f),

hence AA is the projection of the closed set Graph(f)\Op{Graph}(f) along NN\Baire.

(iv) \Rightarrow (iii): Clear, since projections are continuous.

(iv) \Rightarrow (v): NN\Baire is homeomorphic to a Π20\BP{2} subset of 2N\Cant. (Exercise!)

(v) \Rightarrow (vi), (vi) \Rightarrow (iii): Obvious.

The Lusin Separation Theorem

In a course on computability theory one learns that there are effectively inseparable disjoint computably enumerable sets. i.e. disjoint c.e. sets W,ZNW,Z \subseteq \Nat for which no recursive set AA exists with WAW \subseteq A and AZ=A \cap Z = \emptyset.

In contrast to this, disjoint analytic sets can always be separated by a Borel set - they are Borel separable.

Proof

Let f:NNAf:\Baire \to A and g:NNBg:\Baire \to B be continuous surjections.

We argue by contradiction. The key idea is: if AA and BB are Borel inseparable, then, for some i,jNi,j \in \Nat, Ai=f(Ni)A_{\Tup{i}} = f(\Cyl{\Tup{i}}) and Bj=g(Nj)B_{\Tup{j}} = g(\Cyl{\Tup{j}}) are Borel inseparable.

This follows from the following observation:

If the sets Rm,nR_{m,n} separate the sets Pm,QnP_m, \, Q_n (for each m,nm,n), then R=mnRm,nR = \bigcup_m \bigcap_n R_{m,n} separates the sets P=mPm,Q=nQn.P = \bigcup_m P_m, \, Q = \bigcup_n Q_n.

By using this repeatedly, we can construct sequences α,βNN\alpha, \beta \in \Baire such that for all nn, AαnA_{\alpha\Rest{n}} and BβnB_{\beta\Rest{n}} are Borel inseparable, where

Aσ=f(Nσ) and Bσ=g(Nσ).A_{\sigma} = f(\Cyl{\sigma}) \quad \text{ and } \quad B_{\sigma} = g(\Cyl{\sigma}).

Then we have f(α)Af(\alpha) \in A and g(β)Bg(\beta) \in B, and since AA and BB are disjoint, f(α)g(β)f(\alpha) \neq g(\beta). Let U,VU,V be disjoint open sets such that f(α)Uf(\alpha) \in U, g(β)Vg(\beta) \in V. Since ff and gg are continuous, there exists NN such that f(NαN)Uf(\Cyl{\alpha\Rest{N}}) \subseteq U, g(NβN)Vg(\Cyl{\beta\Rest{N}}) \subseteq V, hence UU separates AαNA_{\alpha\Rest{N}} and BβNB_{\beta\Rest{N}}, contradiction.

The Separation Theorem yields a nice characterization of the Borel sets.

Proof

In Theorem 1, chose A=AA = A and B=¬AB = \Co{A}.

It follows from Souslin’s Theorem and the Lusin separation theorem that the analytic sets are not closed under complements.

Sets whose complement is analytic are called co-analytic. Analogous to the levels of the Borel hierarchy, the co-analytic subsets of a Polish space XX are denoted by

Π11(X).\PP{1}(X).

If we define, again analogy to the Borel hierarchy,

Δ11(X)=Σ11(X)Π11(X),\bDelta^1_1(X) = \PS{1}(X) \cap \PP{1}(X),

then Souslin’s Theorem states that

Borel(X)=Δ11(X).\Op{Borel}(X) = \bDelta^1_1(X).

The Souslin operation

Souslin schemes give an alternative presentation of analytic sets which will be useful later.

The analytic sets are precisely the sets that can be obtained by Souslin operations on closed sets. If a Γ\Gamma is a family of subsets of a set XX, we let

AΓ={AP ⁣: P=(Pσ) is a Souslin scheme with PσΓ for all σ}.\mathcal{A}\Gamma = \{\mathcal{A}P \colon \text{ $P = (P_\sigma)$ is a Souslin scheme with $P_\sigma \in \Gamma$ for all $\sigma$} \}.
Proof

Suppose f:NNXf: \Baire \to X is continuous with f(NN)=Af(\Baire) = A. Then

xA    αNN  nN  xf(Nαn).x \in A \iff \exists \alpha \in \Baire \; \forall n \in \Nat \; x \in \Cl{f(\Cyl{\alpha\Rest{n}})}.

Hence if we let Pσ=f(Nσ)P_\sigma = \Cl{f(\Cyl{\sigma})}, then

A=AP,A = \mathcal{A} \,P,

for the Souslin scheme P=(Pσ)P = (P_\sigma).

To see that any set AA in AΠ10(X)\mathcal{A}\,\BP{1}(X) is analytic, consider (**). If the PσP_\sigma are closed, the condition

(α,x)F    nN  xPαn(\alpha,x) \in F \iff \forall n \in \Nat \; x \in P_{\alpha\Rest{n}}

defines a closed subset of NN×X\Baire \times X such that AA is the projection of FF along NN\Baire.

Note that the Souslin scheme (Pσ)(P_\sigma) used in the previous proof has the additional property that

στPσPτ.\sigma \Sleq \tau \quad \Rightarrow \quad P_\sigma \supseteq P_\tau.

Such Souslin schemes are called regular. By replacing any Souslin scheme PσP_\sigma with

Qσ=τσPτ,Q_\sigma = \bigcap_{\tau \Sleq \sigma} P_\tau,

we obtain a regular Souslin scheme Q=(Qσ)Q = (Q_\sigma) with AQ=AP\mathcal{A} \, Q = \mathcal{A}\, P. Moreover, if the PσP_\sigma are from a class Γ\Gamma, and Γ\Gamma is closed under finite intersections, then the QσQ_\sigma are also from Γ\Gamma. In particular, any analytic set can be obtained from a regular Souslin scheme of closed sets via the Souslin operation.