scalar_centered_sampling_qmoment_bernstein_estimate
ProvedThe scalar Bernstein -th moment estimate for the centered-sampling coefficient statistic on the finite Bernoulli observation measure. The statistic is with , a sum of independent mean-zero terms (independence of coordinate inclusion under the Bernoulli powerset measure). Each term has variance and a.s. bound , so the variance proxy is and the sup bound is . The classical Bernstein/Bennett moment inequality for sums of independent centered bounded random variables gives, with ,
where and . This is exactly the -th moment input consumed by the proved Markov-tail node to produce the scalar centered-sampling Bernstein tail. It isolates the single genuinely analytic core (a Rosenthal/Bernstein moment bound for an independent sum on the bespoke powerset measure), which has no existing Mathlib formalization.
import Definitions.Def_matrix_completion_neumann open MatrixCompletion open scoped BigOperators
theorem scalar_centered_sampling_qmoment_bernstein_estimate :
∃ Cbern cbern : ℝ, 0 < Cbern ∧ 0 < cbern ∧
∀ (β : ℝ), 2 < β →
∀ (n₁ n₂ m : ℕ), 0 < n₁ → 0 < n₂ → m ≤ n₁ * n₂ →
∀ (Coeff : Finset (Fin n₁ × Fin n₂) → ℝ)
(B : Matrix (Fin n₁) (Fin n₂) ℝ)
(entryScale frobScale : ℝ),
(∀ Omega : Finset (Fin n₁ × Fin n₂),
Coeff Omega =
matrixEntrySum
(centeredSamplingFluctuation Omega
((m : ℝ) / ((n₁ : ℝ) * (n₂ : ℝ))) B)) →
entrySupNorm B ≤ entryScale →
frobeniusNorm B ≤ frobScale →
∃ q : ℕ, 1 ≤ q ∧
bernoulliExpectation ((m : ℝ) / ((n₁ : ℝ) * (n₂ : ℝ)))
(fun Omega => |Coeff Omega| ^ q) ≤
(Cbern *
(Real.sqrt
((β * Real.log (↑(max n₁ n₂))) /
((m : ℝ) / ((n₁ : ℝ) * (n₂ : ℝ)))) *
frobScale +
((β * Real.log (↑(max n₁ n₂))) /
((m : ℝ) / ((n₁ : ℝ) * (n₂ : ℝ)))) *
entryScale)) ^ q *
(cbern * Real.rpow (↑(max n₁ n₂)) (-β)) := by sorry