Proof of Lemma A.14, p. 79 — affinity equals one minus half squared Hellinger distance
ProvedStatComplexityDM.Estimation.affinity_eq_one_sub_half_hellingeraffinityhellinger-distanceonline-learningp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-bookp2o-v1
Let and be probability vectors on a finite set . Their Hellinger affinity satisfies
The expectation under of , summed over , has the same value.
This is the normalization identity in the conditional exponential-moment calculation for Lemma A.14.
Formalization Note has no factor of . The support restriction keeps the ratio's denominator positive.
Preamble
import Mathlib import Definitions.Def_FoundationsRL_GeneralDM_Divergences import Definitions.Def_StatComplexityDM_Estimation_Sequential
Formal statement
namespace StatComplexityDM.Estimation
open FoundationsRL.GeneralDM Classical
/-- The affinity identity used in the proof of Lemma A.14, p. 79. -/
theorem affinity_eq_one_sub_half_hellinger {Y : Type*} [Fintype Y]
(P Q : Y → ℝ) (hP : StatComplexityDM.LowerBound.IsDist P) (hQ : StatComplexityDM.LowerBound.IsDist Q) :
(∑ y, Real.sqrt (P y * Q y)) = 1 - hellingerSq P Q / 2 ∧
(∑ y, if 0 < P y then P y * Real.sqrt (Q y / P y) else 0) =
1 - hellingerSq P Q / 2 := by sorry
end StatComplexityDM.Estimation
Source
arXiv:2112.13487v3, Appendix A.3.5, proof of Lemma A.14, p. 79, affinity identity following (103)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.