AFTD Auto-Formalizing Theoretical Domains

mutual_power_divergence_subadditive_imp_quadratic_nonneg

theoremverifiedoriginal

For any positive exponent beta, if the mutual power-divergence information with generator f_beta is subadditive under all finite Markov chains Y <- X -> Z, then beta^2 + beta - 1 >= 0.

Statement

theorem mutual_power_divergence_subadditive_imp_quadratic_nonneg (β : ℝ)
    (hβ0 : 0 < β)
    (hsub : mutual_f_information_subadditive (power_divergence_generator β)) :
    0 ≤ β ^ 2 + β - 1
Related work
arXiv:2610.21231 'On the Subadditivity of Mutual f-Information'
Verified
31 Oct 2026
Axioms
Classical.choiceQuot.soundpropext
Built on 4

Read back from the Lean

For every real number β, if: 1. β > 0, and 2. the function power_divergence_generator β satisfies mutual_f_information_subadditive, meaning that for all finite types X, Y, and Z, all probability distributions pX on X (i.e., pX(x) ≥ 0 for all x and ∑_x pX(x) = 1), and all transition kernels PY from X to Y (PY(x, y) ≥ 0 for all x, y and ∑_y PY(x, y) = 1 for all x) and PZ from X to Z (PZ(x, z) ≥ 0 for all x, z and ∑_z PZ(x, z) = 1 for all x), the mutual f-information with respect to power_divergence_generator β of the joint distribution on X and Y × Z defined by (x, (y, z)) ↦ pX(x) * PY(x, y) * PZ(x, z) is less than or equal to the sum of the mutual f-information of the joint distribution on X and Y defined by (x, y) ↦ pX(x) * PY(x, y) and that of the joint distribution on X and Z defined by (x, z) ↦ pX(x) * PZ(x, z), then 0 ≤ β^2 + β - 1 (that is, β^2 + β - 1 is non-negative).

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 reduces the quadratic inequality to a lower bound on β and then invokes an existing theorem. Because β > 0, β^2 + β - 1 ≥ 0 holds if and only if β is at least the positive root (√5 - 1) / 2 of x^2 + x - 1. The claim then follows immediately from a prior result stating that subadditivity of mutual power-divergence information implies β ≥ (√5 - 1) / 2.

  1. It suffices to show that β ≥ (√5 - 1) / 2.

    For β > 0, the quadratic β^2 + β - 1 is non-negative precisely when β is at least the positive root (√5 - 1) / 2 of the polynomial x^2 + x - 1. Rewriting with golden_ratio_quadratic_nonneg_iff reduces the goal 0 ≤ β^2 + β - 1 to this equivalent lower bound.

    line 6

    uses golden_ratio_quadratic_nonneg_iff

      rw [golden_ratio_quadratic_nonneg_iff β hβ0]
  2. (√5 - 1) / 2 ≤ β, completing the proof.

    The theorem mutual_power_divergence_subadditive_imp_golden_ratio_le directly establishes that subadditivity of the mutual power-divergence information for β > 0 forces β ≥ (√5 - 1) / 2. Providing this lemma with the hypotheses hβ0 (that β > 0) and hsub (subadditivity) closes the goal.

    line 7

    uses mutual_power_divergence_subadditive_imp_golden_ratio_le

      exact mutual_power_divergence_subadditive_imp_golden_ratio_le β hβ0 hsub
golden_ratio_quadratic_nonneg_iff
For any positive real number β, the quadratic β^2 + β - 1 is non-negative if and only if β ≥ (√5 - 1) / 2.
mutual_power_divergence_subadditive_imp_golden_ratio_le
If β > 0 and the mutual power-divergence information with parameter β is subadditive under all finite Markov chains, then β ≥ (√5 - 1) / 2.

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

/-- Subadditivity of mutual power-divergence information implies beta^2 + beta - 1 >= 0. -/theorem mutual_power_divergence_subadditive_imp_quadratic_nonneg (β : ℝ)    (hβ0 : 0 < β)    (hsub : mutual_f_information_subadditive (power_divergence_generator β)) :    0 ≤ β ^ 2 + β - 1 := by  rw [golden_ratio_quadratic_nonneg_iff β hβ0]  exact mutual_power_divergence_subadditive_imp_golden_ratio_le β hβ0 hsub
See it in the graphDiscuss this resultChallenge it