AFTD Auto-Formalizing Theoretical Domains

not_mutual_f_information_subadditive_hellinger

theoremverified

The mutual f-information is not subadditive for the Hellinger generator (power divergence with parameter 1/2).

Statement

theorem not_mutual_f_information_subadditive_hellinger :
    ¬ mutual_f_information_subadditive (power_divergence_generator (1 / 2))
Source
arXiv:2610.21231, Section II
Verified
22 Oct 2026
Axioms
Classical.choiceQuot.soundpropext
Built on 4
  • mutual_f_information_subadditiveA function f is a subadditive generator for mutual f-information if for all finite random variables X, Y, Z forming a Markov chain Y -> X…
  • mutual_f_informationFor a joint probability mass function p_{XY} on finite sets, the mutual f-information is Ĩ_f(X; Y) = D_f(p_{XY} ‖ p_X p_Y) = Σ_{x,y}…
  • mutual_hellinger_information_not_subadditiveSubadditivity fails in the Hellinger case β = 1/2: there are binary random variables X, Y, Z with Y ← X → Z and Ĩ_{f_{1/2}}(X; YZ) >…
  • power_divergence_generatorThe power-divergence generator f_β(x) = (x^β − βx + β − 1)/(β(β − 1)), for β ∉ {0, 1} and x ≥ 0 (β = 1/2 gives 2(√x − 1)², the squared…

Read back from the Lean

The theorem asserts the negation of mutual_f_information_subadditive instantiated with f = power_divergence_generator (1 / 2). Explicitly, it asserts that it is NOT the case that: For all types X, Y, Z equipped with Fintype instances, and all functions pX : X → ℝ, PY : X → Y → ℝ, and PZ : X → Z → ℝ, if: 1. ∀ x, 0 ≤ pX x 2. ∑ x, pX x = 1 3. ∀ x y, 0 ≤ PY x y 4. ∀ x, ∑ y, PY x y = 1 5. ∀ x z, 0 ≤ PZ x z 6. ∀ x, ∑ z, PZ x z = 1 then: mutual_f_information (power_divergence_generator (1 / 2)) (fun x yz => pX x * PY x yz.1 * PZ x yz.2) ≤ mutual_f_information (power_divergence_generator (1 / 2)) (fun x y => pX x * PY x y) + mutual_f_information (power_divergence_generator (1 / 2)) (fun x z => pX x * PZ x z). Equivalently, mutual $f$-information with the power divergence generator at parameter $\beta = 1/2$ (the Hellinger divergence generator) fails to be subadditive under conditional independence: there exist finite types $X, Y, Z$, a distribution $p_X$, and transition kernels $P_Y, P_Z$ such that the joint mutual $f$-information $I_f(X; Y, Z)$ is strictly greater than $I_f(X; Y) + I_f(X; Z)$.

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

The proof proceeds by contradiction. Assuming that mutual $f$-information is subadditive for the Hellinger generator, the proof invokes the counterexample `mutual_hellinger_information_not_subadditive`, which provides explicit distributions on binary alphabets for which subadditivity strictly fails. Specializing the general subadditivity hypothesis to these distributions yields an opposing inequality, resulting in an immediate arithmetic contradiction.

  1. Assuming subadditivity, we obtain probability distributions on binary alphabets that strictly violate the subadditivity inequality.

    We assume for contradiction that mutual $f$-information with the Hellinger generator is subadditive (hypothesis $h$). We then invoke `mutual_hellinger_information_not_subadditive` to extract a marginal distribution $p_X$ and conditional distributions $P_Y, P_Z$ on $\mathrm{Fin}\,2$ whose joint mutual information strictly exceeds the sum of the individual mutual informations: $I_f(X; Y) + I_f(X; Z) < I_f(X; (Y, Z))$.

    lines 4–5

    uses mutual_hellinger_information_not_subadditive

      intro h
      obtain ⟨pX, PY, PZ, hX0, hX1, hY0, hY1, hZ0, hZ1, hlt⟩ := mutual_hellinger_information_not_subadditive
  2. Applying the assumed subadditivity to these distributions yields a direct contradiction.

    Instantiating the general subadditivity hypothesis $h$ on the types $\mathrm{Fin}\,2$ and the specific distributions $p_X, P_Y, P_Z$ produces the subadditivity inequality $I_f(X; (Y, Z)) \le I_f(X; Y) + I_f(X; Z)$. Linear arithmetic (`linarith`) detects that this non-strict inequality directly contradicts the strict reverse inequality established previously, completing the proof by contradiction.

    lines 6–7

      have hle := h (Fin 2) (Fin 2) (Fin 2) pX PY PZ hX0 hX1 hY0 hY1 hZ0 hZ1
      linarith
mutual_hellinger_information_not_subadditive
There exist a probability distribution $p_X$ and conditional probability distributions $P_Y, P_Z$ on the two-element set $\mathrm{Fin}\,2$ such that the sum of their individual mutual $f$-informations is strictly less than their joint mutual $f$-information with respect to the Hellinger generator.

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 generator (power divergence with parameter 1/2) is not subadditive for mutual f-information. -/theorem not_mutual_f_information_subadditive_hellinger :    ¬ mutual_f_information_subadditive (power_divergence_generator (1 / 2)) := by  intro h  obtain ⟨pX, PY, PZ, hX0, hX1, hY0, hY1, hZ0, hZ1, hlt⟩ := mutual_hellinger_information_not_subadditive  have hle := h (Fin 2) (Fin 2) (Fin 2) pX PY PZ hX0 hX1 hY0 hY1 hZ0 hZ1  linarith
See it in the graphDiscuss this resultChallenge it