dlp_bernoulli_three_copy_triangle_scalar
Provedde la Peña–Montgomery-Smith 1995 (arXiv:math/9309211) Lemma 1, transported to the finite Bernoulli powerset probability model. For any scalar statistic Z on the sampling set and any threshold t: P(t ≤ |Z(Ω)|) ≤ 3·P(2t/3 ≤ |Z(Ω₁)+Z(Ω₂)|), where the LHS is the single-copy bernoulliEventProb and the RHS is the decoupled three-copy bernoulliTripleEventProb (the third copy does not appear in the event). Stated with absolute values on both sides to faithfully match Lemma 1 at the scalar instance V=ℝ (where ‖r‖=|r|); the signed/ordered variant is false because the union-cover in Lemma 1's proof genuinely needs the norm/triangle inequality 2X=(X+Y)+(X+Z)−(Y+Z). This is the matrix→scalar 3-copy desymmetrization triangle on the bespoke powerset model: applied at the dual image Z=⟨M·,x,y⟩ it supplies the Lemma-1 step of the de la Peña forward decoupling bound on the powerset measure.
import Definitions.Def_matrix_completion_bernoulli import Definitions.Def_matrix_completion_neumann import Definitions.Def_matrix_completion_bernoulli_measure open MatrixCompletion open scoped BigOperators Classical ENNReal open MeasureTheory ProbabilityTheory
theorem dlp_bernoulli_three_copy_triangle_scalar
{n1 n2 : Nat} (p : NNReal) (hp : p ≤ 1)
(Z : Finset (Fin n1 × Fin n2) → ℝ) (t : ℝ) :
bernoulliEventProb (p : ℝ) (fun Omega => t ≤ |Z Omega|)
≤ 3 * bernoulliTripleEventProb (p : ℝ)
(fun Omega1 Omega2 _Omega3 => 2 * t / 3 ≤ |Z Omega1 + Z Omega2|) := by
sorry