AFTD Auto-Formalizing Theoretical Domains

qa_pool_tsallis_three_unique

theoremverified

For the cubic Tsallis rule the QA pool of any forecasts and weights is unique.

Statement

theorem qa_pool_tsallis_three_unique {n m : ℕ} (ps : Fin m → Fin n → ℝ) (w : Fin m → ℝ)
    (x x' : Fin n → ℝ)
    (hx : is_qa_pool (tsallis_expected_reward 3) (tsallis_exposure 3) ps w x)
    (hx' : is_qa_pool (tsallis_expected_reward 3) (tsallis_exposure 3) ps w x') : x = x'
Source
step towards tsallis_three_qa_weight_score_not_concave (concavity without convex exposure, open in arXiv:2102.07081 v2, Sec. 7)
Verified
06 Oct 2026
Axioms
Classical.choiceQuot.soundpropext
Built on 3
  • is_qa_poolx is the quasi-arithmetic pool of forecasts p_i with weights w_i: x lies in the simplex and minimizes G(y) - <y, sum_i w_i g(p_i)> over…
  • tsallis_expected_rewardThe Tsallis expected reward with natural parameter gamma: the sum of x_k^gamma.
  • tsallis_exposureThe Tsallis exposure (gradient of the expected reward): gamma x_k^(gamma - 1).

Lean source view module on GitHub

/-- The QA pool for the cubic Tsallis rule is unique. -/theorem qa_pool_tsallis_three_unique {n m : ℕ} (ps : Fin m → Fin n → ℝ) (w : Fin m → ℝ)    (x x' : Fin n → ℝ)    (hx : is_qa_pool (tsallis_expected_reward 3) (tsallis_exposure 3) ps w x)    (hx' : is_qa_pool (tsallis_expected_reward 3) (tsallis_exposure 3) ps w x') : x = x' := by  obtain ⟨⟨hx0, hx1⟩, hxmin⟩ := hx  obtain ⟨⟨hx'0, hx'1⟩, hx'min⟩ := hx'  set V : Fin n → ℝ := fun k => ∑ i, w i * tsallis_exposure 3 (ps i) k  have hm : (fun k => (x k + x' k) / 2) ∈ stdSimplex ℝ (Fin n) := by    refine ⟨fun k => by have := hx0 k; have := hx'0 k; positivity, ?_⟩    rw [← Finset.sum_div, Finset.sum_add_distrib, hx1, hx'1]; norm_num  have h1 := hxmin _ hm  have h2 := hx'min _ ⟨hx0, hx1⟩  have h3 := hxmin _ ⟨hx'0, hx'1⟩  unfold tsallis_expected_reward at h1 h2 h3  -- the convexity gap of the cube at the midpoint  have key : ∑ k, (3 / 8 : ℝ) * (x k + x' k) * (x k - x' k) ^ 2 ≤ 0 := by    have e : ∀ k, (3 / 8 : ℝ) * (x k + x' k) * (x k - x' k) ^ 2 =        (x k ^ 3 + x' k ^ 3) / 2 - ((x k + x' k) / 2) ^ 3 := by intro k; ring    simp_rw [e]    rw [Finset.sum_sub_distrib, ← Finset.sum_div, Finset.sum_add_distrib]    have lin : ∑ k, (x k + x' k) / 2 * V k = (∑ k, x k * V k + ∑ k, x' k * V k) / 2 := by      rw [← Finset.sum_add_distrib, Finset.sum_div]; congr 1; ext k; ring    simp only [V] at lin    linarith  have hterm : ∀ k ∈ Finset.univ, 0 ≤ (3 / 8 : ℝ) * (x k + x' k) * (x k - x' k) ^ 2 := by    intro k _; have := hx0 k; have := hx'0 k; positivity  have hz := (Finset.sum_eq_zero_iff_of_nonneg hterm).mp (le_antisymm key (Finset.sum_nonneg hterm))  funext k  have := hz k (Finset.mem_univ k)  rcases mul_eq_zero.mp this with h | h  · have := hx0 k; have := hx'0 k    have : x k + x' k = 0 := by linarith    linarith  · nlinarith [pow_eq_zero_iff (n := 2) (two_ne_zero) |>.mp h]
See it in the graphDiscuss this resultChallenge it