AFTD Auto-Formalizing Theoretical Domains

qa_pool_tsallis_three_of_kkt

lemmaverified

For the two counterexample forecasts and any weights, a simplex point meeting the KKT conditions is the cubic Tsallis QA pool.

Statement

lemma qa_pool_tsallis_three_of_kkt (w : Fin 2 → ℝ) (x : Fin 3 → ℝ) (lam : ℝ)
    (hx : x ∈ stdSimplex ℝ (Fin 3))
    (hkkt : ∀ k, lam ≤ 3 * x k ^ 2 - ∑ i, w i * tsallis_exposure 3 (qa_counterexample_forecasts i) k ∧
      (0 < x k → 3 * x k ^ 2 - ∑ i, w i * tsallis_exposure 3 (qa_counterexample_forecasts i) k = lam)) :
    is_qa_pool (tsallis_expected_reward 3) (tsallis_exposure 3) qa_counterexample_forecasts w 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 5
  • 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…
  • qa_counterexample_forecastsThe two expert forecasts (0, 0, 1) and (1/4, 3/4, 0) on three outcomes.
  • qa_pool_tsallis_three_auxKKT sufficiency on the 3-simplex for sum x_k^3 - <x, V>: if 3 x_k^2 - V_k >= lam for all k with equality where x_k > 0, then x minimizes…
  • 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).
Used by 1

Lean source view module on GitHub

lemma qa_pool_tsallis_three_of_kkt (w : Fin 2 → ℝ) (x : Fin 3 → ℝ) (lam : ℝ)
    (hx : x ∈ stdSimplex ℝ (Fin 3))
    (hkkt : ∀ k, lam ≤ 3 * x k ^ 2 - ∑ i, w i * tsallis_exposure 3 (qa_counterexample_forecasts i) k ∧
      (0 < x k → 3 * x k ^ 2 - ∑ i, w i * tsallis_exposure 3 (qa_counterexample_forecasts i) k = lam)) :
    is_qa_pool (tsallis_expected_reward 3) (tsallis_exposure 3) qa_counterexample_forecasts w x :=
  ⟨hx, qa_pool_tsallis_three_aux _ x lam hx hkkt⟩
See it in the graphDiscuss this resultChallenge it