Proof of Theorem 3.13, p. 14 — x ∈ X^{N,2t}_{ε,(δ−β)} and ‖x − z‖∞ ≤ t/L give F^N_ε(−t; z) ≥ 1 − δ + β
ProvedSmoothCCP.Feasibility.transfer_stepLet be built from an admissible and fix sample values . Let , , and suppose is -Lipschitz in the sup norm at every drawn scenario:
If , i.e. and , and satisfies , then
This is the step that transfers sample feasibility from an arbitrary point to its representative in the net, at the cost of halving the shift from to .
Formalization Note The paper's argument holds "w.p.1": Assumption 3.12 holds for almost every , hence almost surely at all drawn scenarios at once. This item is the deterministic core on one realisation of the sample, with Assumption 3.12 stated at the drawn points; the passage from almost-sure Lipschitz continuity to the drawn points is part of the proof of Theorem 3.13. Monotonicity of is not assumed; it follows from admissibility.
import Mathlib import Definitions.Def_SmoothCCP_Feasibility_Setting open MeasureTheory
namespace SmoothCCP.Feasibility
theorem transfer_step {n N : ℕ} {Ξ : Type} (C : (Fin n → ℝ) → Ξ → ℝ) (X : Set (Fin n → ℝ))
(ε : ℝ) (γ : ℝ → ℝ) (hγ : AdmissibleGamma ε γ) (ξs : Fin N → Ξ)
(L t δ β : ℝ) (hL : 0 < L) (ht : 0 < t)
(hLip : ∀ i, ∀ x ∈ X, ∀ y ∈ X, |C x (ξs i) - C y (ξs i)| ≤ L * ‖x - y‖)
(x z : Fin n → ℝ) (hx : x ∈ sampleFeasible C X ε γ ξs (2 * t) (δ - β)) (hz : z ∈ X)
(hxz : ‖x - z‖ ≤ t / L) :
1 - δ + β ≤ sampleCdf C ε γ ξs (-t) z := by sorry
end SmoothCCP.Feasibility
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.