Product weights certify an upper bound on the entropy penalty
Provedmme_entropy_penalty_le_product_dualmatrix-multiplicationregional-extraction
A feasible positive product of coordinate weights bounds the maximum-entropy penalty in natural-log units. This gives a certificate interface without requiring an exact entropy maximizer. Numerical feasibility and logarithm bounds for released recipes remain separate obligations. The full matrix exponent bound remains a separate obligation.
Preamble
import Theorems.Thm_mme_same_marginal_entropy_le_product_dual import Mathlib.Tactic.FieldSimp import Mathlib.Tactic.Ring open scoped BigOperators open MME.RegionRate MME.RecursiveThinSplit
Formal statement
theorem mme_entropy_penalty_le_product_dual
{half : ℕ} {parent : Fin 3 → ℕ} (alpha : Split half parent → ℝ)
(ha : ∀ c, 0 ≤ alpha c) (hprob : ∑ c, alpha c = 1)
(weights : Fin 3 → Fin (half + 1) → ℝ)
(hpos : ∀ i j, 0 < weights i j)
(hmass : ∑ c : Split half parent, ∏ i, weights i (c.val i) ≤ 1) :
Real.log 2 * entropyPenalty alpha ≤
-(∑ i, ∑ j, mme_modern_marginal (fun c : Split half parent => c.val i) alpha j *
Real.log (weights i j)) - entropy alpha := by sorrySource
Kernel-checked regional count identities and exact extraction with empty regions.