AFTD Auto-Formalizing Theoretical Domains

hellinger_mutual_information_of_independent

theoremverified

For independent finite random variables U and V with product distribution p(u, v) = a(u) * b(v) where marginal distributions a and b each sum to 1, the Hellinger mutual information hellinger_mutual_information (fun u v => a u * b v) is zero.

Statement

theorem hellinger_mutual_information_of_independent {U V : Type*} [Fintype U] [Fintype V]
    (a : U → ℝ) (b : V → ℝ) (ha : ∑ u, a u = 1) (hb : ∑ v, b v = 1) :
    hellinger_mutual_information (fun u v => a u * b v) = 0
Source
folklore
Verified
13 Nov 2026
Axioms
Classical.choiceQuot.soundpropext
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

For any finite types $U$ and $V$, and any functions $a : U \to \mathbb{R}$ and $b : V \to \mathbb{R}$ satisfying $\sum_{u \in U} a(u) = 1$ and $\sum_{v \in V} b(v) = 1$, the Hellinger mutual information of the product function $(u, v) \mapsto a(u) b(v)$ is equal to $0$.

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

The proof, explained

By definition, the Hellinger mutual information is a double sum of squared differences (√p(u, v) - √(p_U(u) * p_V(v)))^2 over all pairs (u, v). For independent random variables with product distribution p(u, v) = a(u) * b(v), the marginal sums factor into a(u) * ∑ v', b(v') and b(v) * ∑ u', a(u'), which equal a(u) and b(v) respectively because the marginals each sum to 1. Consequently, the two square-root terms are identical for every pair (u, v), so every term in the sum is zero.

  1. To show the Hellinger mutual information is zero, it suffices to prove that each summand (√(a u * b v) - √((∑ v', a u * b v') * ∑ u', a u' * b v))^2 vanishes for arbitrary u and v.

    Unfolding the definition of Hellinger mutual information expresses it as the double sum ∑ u, ∑ v, (√(a u * b v) - √((∑ v', a u * b v') * ∑ u', a u' * b v))^2. Applying Finset.sum_eq_zero to each sum reduces the problem to showing that the individual summand is zero for arbitrary elements u ∈ U and v ∈ V.

    lines 5–9

    uses Finset.sum_eq_zero

      unfold hellinger_mutual_information
      apply Finset.sum_eq_zero
      intro u hu
      apply Finset.sum_eq_zero
      intro v hv
  2. The marginal sum over V satisfies ∑ v', a u * b v' = a u.

    Factoring the constant a u out of the sum using the distributive law (Finset.mul_sum) gives a u * ∑ v', b v'. Since ∑ v', b v' = 1 by assumption hb, this simplifies to a u * 1 = a u.

    lines 10–12

    uses Finset.mul_sum

      have h1 : (∑ v', a u * b v') = a u := by
        rw [← Finset.mul_sum]
        simp [hb]
  3. The marginal sum over U satisfies ∑ u', a u' * b v = b v.

    Factoring the constant b v out of the sum using the distributive law (Finset.sum_mul) gives (∑ u', a u') * b v. Since ∑ u', a u' = 1 by assumption ha, this simplifies to 1 * b v = b v.

    lines 13–15

    uses Finset.sum_mul

      have h2 : (∑ u', a u' * b v) = b v := by
        rw [← Finset.sum_mul]
        simp [ha]
  4. The summand evaluates to zero, completing the proof.

    Substituting the marginal identities h1 and h2 into the summand yields (√(a u * b v) - √(a u * b v))^2. By sub_self, the difference inside the square is zero, leaving 0^2 = 0.

    lines 16–17

    uses sub_self

      rw [h1, h2, sub_self]
      simp
Finset.sum_eq_zero
A finite sum is zero if every term in the sum is zero.
Finset.mul_sum
Multiplication distributes over finite summation: a * ∑ i, f i = ∑ i, a * f i.
Finset.sum_mul
Multiplication distributes over finite summation on the right: (∑ i, f i) * a = ∑ i, f i * a.
sub_self
Subtracting any element from itself gives zero: a - a = 0.

Written by a model (gemini-3.8-flash) from the Lean proof and the goal states Lean reported at each step. Lean verified the proof; this explanation is a reading aid and is not itself machine-checked.

Lean source view module on GitHub

/-- The Hellinger mutual information of independent random variables is zero. -/theorem hellinger_mutual_information_of_independent {U V : Type*} [Fintype U] [Fintype V]    (a : U → ℝ) (b : V → ℝ) (ha : ∑ u, a u = 1) (hb : ∑ v, b v = 1) :    hellinger_mutual_information (fun u v => a u * b v) = 0 := by  unfold hellinger_mutual_information  apply Finset.sum_eq_zero  intro u hu  apply Finset.sum_eq_zero  intro v hv  have h1 : (∑ v', a u * b v') = a u := by    rw [← Finset.mul_sum]    simp [hb]  have h2 : (∑ u', a u' * b v) = b v := by    rw [← Finset.sum_mul]    simp [ha]  rw [h1, h2, sub_self]  simp
See it in the graphDiscuss this resultChallenge it