Alphabeta Math
LemmaStatement: AI-adaptedProof: AI-generatedPipeline-generatedjudge pass (gpt-6-luna)
How statement and proof provenance work

The first chip identifies the source of the statement or construction; the second identifies the source of its local proof or verification.

  • Literature-sourced: the exact statement appears in a cited source; only wording and notation differ.
  • AI-adapted: a semantically identical restatement of literature-sourced material, modulo indexing, notation, and boundary cases adopted by the library.
  • AI-generated: a genuinely novel statement formulated by AI, with no source for the claim itself.

These labels describe origin, not correctness: citations and verification chips remain separate evidence.

A finite label has a conditional probability function

Statement

Assume AC. Let a supervised prediction problem have a measurable instance space (X,ΣX), finite label set Y={y1,…,yK} with distinct labels, the full power-set sigma-algebra on Y, and K≥2, and joint probability law P on X×Y; let X and Y be its coordinate random variables and let G={X−1(A):A∈ΣX}, the sigma-algebra generated by X. There is a measurable function η=(η1,…,ηK):X→ΔK, where ΔK={p∈[0,1]K:∑jpj=1}⊂RK has its subspace Borel sigma-algebra, such that for every label index j and A∈ΣX, P(X∈A, Y=yj)=E[ηj(X)1{X∈A}].

Statement references

Supervised prediction problems: supplies measurable instance and label spaces, the joint law, and the coordinate random variables; here the finite label sigma-algebra is explicitly the full power set.

Conditional expectation given a sigma algebra: a conditional-expectation version is measurable with respect to the conditioning sigma-algebra and has the same integrals on its events.

Conditional expectation exists by radon nikodym: under AC, an integrable real random variable has a conditional-expectation version given any sub-sigma-algebra.

Durrett, Section 4.1: defines conditional expectation by event integrals and constructs it from Radon–Nikodym derivatives.

Facts & Assumptions

Given: AC; a finite-label supervised prediction problem with joint law P and full power-set label sigma-algebra; the instance coordinate X; and the sub-sigma-algebra G={X−1(A):A∈ΣX}.

[A1]

Assume AC, the choice principle used by the published conditional-expectation existence theorem. The Axiom of Choice

[F1]

Each indicator 1{Y=yj} is integrable because it is bounded by 1 on a probability space; under AC, its conditional-expectation version exists and has the defining event integrals. Probability measures and probability spaces Conditional expectation given a sigma algebra Conditional expectation exists by radon nikodym

[F2]

Conditional expectation preserves order and constants almost surely. Basic algebra and order properties of conditional expectation

[F6]

A conditional-expectation version is unique almost surely. Conditional expectation is unique almost surely

[F3]

Integrable functions equal almost surely have identical integrals over every measurable event. Two integrable functions are equal almost everywhere exactly when all of their indefinite integrals agree

[F4]

Measurability is determined by preimages of measurable sets. A measurable function between measurable spaces

[F5]

The family of all inverse images X−1(A) is a sigma-algebra because inverse images preserve complements and countable unions; by definition it is σ(X).

[F7]

Finite sums, products by indicators, and level sets of measurable real functions are measurable. Closure properties of measurable functions used by the integral A measurable function between measurable spaces

Proof

1.1A1F1F2F3F6F7given

For each j, choose a version Zj of E[1{Y=yj}∣G], which exists by [A1,F1]. Order preservation and preservation of constants give 0≤Zj≤1 almost surely by [F2]. Set Nj={Zj<0}∪{Zj>1}∈G and define Zj′ to equal Zj on Njc and 0 on Nj. Since both pieces are G-measurable, Zj′ is G-measurable; it takes values in [0,1] everywhere and equals Zj almost surely. By [F3], it has the same event integrals, so it remains a conditional-expectation version. Linearity shows that ∑jZj is a version of the conditional expectation of ∑j1{Y=yj}=1; the constant 1 is another version, so [F6] gives ∑jZj=1 almost surely. Hence ∑jZj′=1 almost surely.

2.1A1F4F5F7step 1.1chooseconstruct

For each label j and rational q∈Q∩[0,1], the event {Zj′<q}∈G equals X−1(Aj,q) for some Aj,q∈ΣX by [F5]. AC permits choosing these sets for the countable family of label-rational pairs. Define η~j(x)=inf⁡({q∈Q∩[0,1]:x∈Aj,q}∪{1}). For 0<t≤1, {x:η~j(x)<t}=⋃q<tAj,q; for t≤0 this sublevel set is empty and for t>1 it is all of X. Thus each η~j is measurable and takes values in [0,1] by [F4,F7].

3.1F7step 1.1step 2.1construct

For every j and ω, the rationals q satisfying X(ω)∈Aj,q are exactly those with Zj′(ω)<q. Their infimum is Zj′(ω) when this value is below 1; at value 1 the rational set is empty and the filler 1 gives the same value. Thus η~j(X)=Zj′ everywhere. The measurable set D={x:∑jη~j(x)=1} is well-defined by [F7]. Define η(x)=(η~1(x),…,η~K(x)) on D and η(x)=(1,0,…,0) off D. Its coordinates are measurable. The inverse image of an open rational box in RK is a finite intersection of coordinate preimages, and these boxes form a countable basis, so η is Borel measurable into RK and hence measurable into the subspace ΔK. It takes values in ΔK everywhere, and ηj(X)=Zj′ almost surely because ∑jZj′=1 almost surely.

4.1F1F3step 1.1step 3.1given∎

For every A∈ΣX and label index j, the event {X∈A} belongs to G. The conditional-expectation identity [F1] and almost-sure equality [F3] therefore give P(X∈A,Y=yj)=E[1{X∈A}Zj]=E[1{X∈A}Zj′]=E[1{X∈A}ηj(X)]. This is the claimed event identity for every class.

Depends on

Used by

Dependency tree · two levels

31 results within two dependency steps of this one, each drawn at its shortest distance from it. An arrow runs from a result to what uses it, so the chart reads left to right and ends at this result, which carries a heavier outline. Every node is a link to that result. Click elsewhere on the chart to enlarge it.

Sources