AFTD Auto-Formalizing Theoretical Domains

hellinger_observation_factor

def

A real number C satisfies the m-observation multiplicative subadditivity inequality for Hellinger mutual information if for every finite type X and family of finite observation types Y_i (for i in Fin m), every prior distribution pX on X and conditionally independent channels P_i from X to Y_i, the joint Hellinger mutual information H(X; Y_1, ..., Y_m) is at most C times the sum of the individual Hellinger mutual informations sum_i H(X; Y_i).

Statement

def hellinger_observation_factor (m : ℕ) (C : ℝ) : Prop
Defined in
arXiv:2610.21231 (On the Subadditivity of Mutual f-Information)
Verified
13 Nov 2026
Axioms
none listed
Built on 1
  • hellinger_mutual_informationFor finite types U and V, the Hellinger mutual information of a joint distribution p on U × V is defined as ∑ u, ∑ v, (Real.sqrt (p u v) -…

Read back from the Lean

Given a natural number m and a real number C, the proposition hellinger_observation_factor m C asserts that: For every finite type X and every family of finite types Y : Fin m → Type, and for every function pX : X → ℝ and family of functions P : ∀ i : Fin m, X → Y i → ℝ, if: 1. pX is non-negative: ∀ x : X, 0 ≤ pX x, 2. pX sums to 1: ∑ x : X, pX x = 1, 3. P is non-negative: ∀ (i : Fin m) (x : X) (y : Y i), 0 ≤ P i x y, and 4. for each i and x, P i x sums to 1: ∀ (i : Fin m) (x : X), ∑ y : Y i, P i x y = 1, then the Hellinger mutual information between X and the product observation (y : (i : Fin m) → Y i) under the conditionally independent joint distribution (x, y) ↦ pX x * ∏ i, P i x (y i) is bounded above by C times the sum over i : Fin m of the Hellinger mutual information between X and Y i under the joint distribution (x, y) ↦ pX x * P i x y.

Written by a model that saw only the Lean, never the English above. If the two disagree, that is worth a challenge.

Lean source view module on GitHub

/-- Multiplicative factor for Hellinger mutual information with m conditionally independent observations -/
def hellinger_observation_factor (m : ℕ) (C : ℝ) : Prop := ∀ (X : Type) (Y : Fin m → Type) [Fintype X] [∀ i, Fintype (Y i)]
    (pX : X → ℝ) (P : ∀ i, X → Y i → ℝ),
    (∀ x, 0 ≤ pX x) → (∑ x, pX x = 1) →
    (∀ i x y, 0 ≤ P i x y) → (∀ i x, ∑ y, P i x y = 1) →
    hellinger_mutual_information (fun x (y : (i : Fin m) → Y i) => pX x * ∏ i, P i x (y i)) ≤
      C * ∑ i, hellinger_mutual_information (fun x y => pX x * P i x y)
See it in the graphDiscuss this resultChallenge it