Alphabeta Math
TheoremStatement: Literature-sourcedProof: AI-adaptedPipeline-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.

Hellinger invariance and comparison with library total variation

Statement

Assume AC for the imported density-representative convention. Let P,Q be probability laws on a measurable space, both dominated by one specified common sigma-finite positive measure. Then the root-L2 Hellinger distance from Root-L2 Hellinger distance for dominated laws is independent of the choice of common dominating measure. With the library convention ∥P−Q∥TV:=sup⁡A∣P(A)−Q(A)∣ from Total variation distance for probability laws,

∥P−Q∥TV≤h(P,Q),h(P,Q)2≤2∥P−Q∥TV.

Facts & Assumptions

Given: Probability laws P,Q with a specified common sigma-finite dominating measure. The conclusions are the common-measure invariance and both displayed inequalities.

[A1]

Assume the Axiom of Choice for the general density-selection convention in Likelihood densities. After finite-valued density representatives are supplied, the calculations use no further choice. The Axiom of Choice

[F1]

For finite-valued nonnegative density representatives p,q against a common measure μ, h(P,Q)2=∫(p−q)2 dμ. Root-L2 Hellinger distance for dominated laws

[F2]

The library total-variation distance is sup⁡A∣P(A)−Q(A)∣, with no factor 1/2 in the event supremum. Total variation distance for probability laws

[F3]

A common dominating measure is sigma-finite and satisfies P,Q≪μ. Dominated statistical experiments

[F4]

Under AC, a law dominated by the specified sigma-finite measure has a finite-valued nonnegative measurable density representative. Likelihood densities

[F5]

For square-integrable real functions f,g, ∫∣fg∣ dμ≤∥f∥2∥g∥2. Holder's inequality for integrals, including the endpoint cases

[F6]

A finite measure is sigma-finite, and a sigma-finite measure has a countable finite-measure exhaustion. Finite, sigma-finite, and semifinite measures

[F7]

Sums, products, scalar multiples, and lattice operations of finite real-valued measurable functions are measurable. Arithmetic and lattice operations preserve measurability whenever they are defined

[F8]

A set defined by a real threshold of a measurable real-valued function is measurable. Threshold characterisations of real-valued and extended-real-valued measurability

[F9]

For nonnegative measurable f,g, the density measure satisfies ∫g d(f dμ)=∫gf dμ. Integrating against a density agrees with integrating the product

Proof

Given: The hypotheses in the Statement.

1.1A1F3F4F6F7F10given

By [A1], [F3], and [F4], take finite-valued nonnegative densities p=dP/dμ and q=dQ/dμ for a common sigma-finite dominator μ. Define ρ(A):=P(A)+Q(A). For disjoint measurable Ak, countable additivity of P,Q and regrouping nonnegative series give ρ(⋃kAk)=∑kρ(Ak); also ρ(∅)=0 and ρ(X)=2. Thus ρ is a finite, hence sigma-finite, measure by [F6]. Additivity of the nonnegative integral [F10] gives ρ(A)=∫A(p+q) dμ, with p+q measurable by [F7].

1.2F2F3F4F7F8F10algebra

Using p,q against μ, the set B:={p≥q} is measurable by [F7,F8]. On B, p=(p−q)+q, so additivity [F10] gives δ:=P(B)−Q(B)=∫B(p−q) dμ≥0. On Bc, q=(q−p)+p and the total masses are one, hence ∫Bc(q−p) dμ=Q(Bc)−P(Bc)=δ. Additivity then gives ∫∣p−q∣ dμ=2δ. For any measurable A, the difference on A∩B lies in [0,δ] by monotonicity [F10], while the difference on A∩Bc lies in [−δ,0]; thus ∣P(A)−Q(A)∣≤δ, and A=B attains δ. Therefore [F2] gives ∥P−Q∥TV=δ=12∫∣p−q∣ dμ.

2.1F7F8F9step 1.1algebra

On {p+q>0} define pˉ=p/(p+q) and qˉ=q/(p+q), and set both to zero on {p+q=0}. For r<0, {pˉ>r} is the whole space; for r≥0, it is {p+q>0}∩{p−r(p+q)>0}. Thus [F7,F8] make pˉ measurable, and the same argument applies to qˉ. Since p=q=0 where p+q=0, [F9] gives ∫Apˉ dρ=∫Apˉ(p+q) dμ=P(A) and ∫Aqˉ dρ=Q(A) for every measurable A. Hence pˉ,qˉ are finite-valued density representatives of P,Q against ρ.

2.2F1F5F10step 1.2algebra

Pointwise ∣p−q∣=∣p−q∣(p+q). Hölder [F5] therefore gives ∫∣p−q∣ dμ≤h(P,Q)(∫(p+q)2 dμ)1/2. The squared second factor is 2+2∫pq dμ≤4, since [F5] applied to p,q bounds their product integral by 1. Step 1.2 now gives ∥P−Q∥TV≤h(P,Q).

3.1F1F7F8F9F10step 1.1step 2.1algebra

Write Hλ(P,Q)2 for the defining Hellinger integral computed from supplied densities against a common dominator λ. Applying [F1] and [F9] to the densities in step 2.1 gives Hρ(P,Q)2=∫(pˉ−qˉ)2 dρ=∫{p+q>0}(p−q)2 dμ=Hμ(P,Q)2, because both original densities vanish where p+q=0. For any measure α and finite-valued nonnegative densities f,g of the same finite measure Λ with respect to α, they agree α-almost everywhere: on Em:={f≥g+1/m}, equality of their density integrals gives 0=∫Em(f−g) dα≥α(Em)/m, so α(Em)=0. Since {f>g}=⋃m≥1Em, and the same argument with f,g swapped handles {g>f}, this proves uniqueness almost everywhere. Thus changing supplied representatives against a fixed dominator does not change the Hellinger integral.

4.1step 1.1step 2.1step 3.1F1F3F4

For any other common sigma-finite dominator ν, repeat steps 1.1 and 2.1 using its densities to obtain density representatives pˉν,qˉν against the same finite measure ρ=P+Q. The uniqueness argument in step 3.1 gives pˉν=pˉ and qˉν=qˉ ρ-almost everywhere. The same density-change calculation as in step 3.1 therefore identifies the Hellinger integral against ν with the same canonical value against ρ, and hence with that against μ.

5.1F1F10step 1.2algebra∎

Pointwise (p−q)2≤∣p−q∣. Integrating by monotonicity [F10] and using [F1] and step 1.2 gives h(P,Q)2≤∫∣p−q∣ dμ=2∥P−Q∥TV.

Depends on

Used by

Dependency tree · two levels

48 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