§4.8, classical Hellinger identities (34)–(37)
OpenQAlgorithms.Papers.Independent.Intakenext100.P45.classicalHellingerLet be finite probability laws with identical support. Define , , and . Then is a probability law and, for ,
A parameter minimizes exactly when these two relative entropies coincide. A minimizer exists and satisfies . It is unique when ; for equal laws every parameter minimizes. Logs concern the common positive support; ambient zero coordinates are allowed.
import Definitions.Def_QAlgorithms_Papers_Independent_Intakenext100_P45
noncomputable section
namespace QAlgorithms.Papers.Independent.Intakenext100.P45
open QAlgorithms Filter
open scoped Matrix ComplexConjugate ComplexOrder ENNReal BigOperators Topology
open QAlgorithms.Papers.Independent.Information100.P07
(support positivePart positiveProjection IsProjection)
open QAlgorithms.Papers.Independent.Information100.P56 (IsLaw)
open QAlgorithms.Papers.Independent.Information100.P60 (Mat copies)
open QAlgorithms.Papers.Independent.Dependency50.P28 (logNonnegativeExtended relativeEntropy)
open QAlgorithms.Papers.Independent.Claimed50.P15 (positiveMatrixRpow)
attribute [local instance] Classical.propDecidable Classical.decEq
variable {I J : Type} [Fintype I] [DecidableEq I] [Fintype J] [DecidableEq J]
/-- arXiv:0708.4282v1, PDF p.19: §4.8, classical Hellinger identities (34)–(37). Source correction: PDF p.19 claims unconditional uniqueness; uniqueness is restricted to p != q, since equal laws have every s as a minimizer. Independent moderation required. -/
theorem classicalHellinger (p q : I → ℝ) (hp : IsLaw p) (hq : IsLaw q) (heq : SamePositiveSupport p q) :
(∀ s ∈ Set.Icc (0 : ℝ) 1,
IsLaw (hellingerArc p q s) ∧
kl (hellingerArc p q s) p =
((s*logDerivative p q s-Real.log (classicalOverlap p q s) : ℝ) : EReal) ∧
kl (hellingerArc p q s) q =
((-(1-s)*logDerivative p q s-Real.log (classicalOverlap p q s) : ℝ) : EReal) ∧
((∀ t ∈ Set.Icc (0 : ℝ) 1, classicalOverlap p q s ≤ classicalOverlap p q t) ↔
kl (hellingerArc p q s) p = kl (hellingerArc p q s) q)) ∧
(∃ s ∈ Set.Icc (0 : ℝ) 1,
(∀ t ∈ Set.Icc (0 : ℝ) 1, classicalOverlap p q s ≤ classicalOverlap p q t) ∧
kl (hellingerArc p q s) p = classicalDistance p q ∧
kl (hellingerArc p q s) q = classicalDistance p q) ∧
(p ≠ q → ∃! s : ℝ, s ∈ Set.Icc (0 : ℝ) 1 ∧
kl (hellingerArc p q s) p = kl (hellingerArc p q s) q) := by sorry
end QAlgorithms.Papers.Independent.Intakenext100.P45
end
Read-back
What the Lean code literally says, in plain math · gpt-6.1-sol
For every finite index type with decidable equality, let satisfy , , and if and only if for every . Define
where real powers and division use their total conventions, for and otherwise, and the derivative is the ordinary derivative of the function on , including when evaluated at or . For real functions on the same index type, write if some has , and otherwise write , viewed as an extended real number. For every , is nonnegative and sums to , and
moreover, for every if and only if . There exists minimizing on this interval for which both these divergences equal . If , there is exactly one real satisfying both and . Common zero coordinates are allowed, but disjoint positive supports and an empty are excluded by the hypotheses; no uniqueness is asserted when .
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.