Corollary 1, p. 13 — the robust Nadaraya–Watson budget equals the perspective program over (s, P)
ProvedBootRobust.NWDual.corollary_1Let be any function assigning an extended real number to pairs of distributions on , let be a reference distribution, a radius, positive weights and the losses of a fixed decision . Then the robust Nadaraya–Watson budget (24) with ,
equals
where ranges over . Both sides are extended reals (an empty feasible set gives ).
The reformulation turns a ratio objective into a linear one; when is convex in its first argument the constraint is jointly convex, which is how the paper obtains a convex program.
Formalization Note The statement is the equality of optimal values for an arbitrary ; the convexity claimed by the corollary's wording ("can be reformulated as the convex optimization problem") is not part of it. "" is written explicitly because the page's is defined only on .
import Mathlib import Definitions.Def_BootRobust_NWDual_Setting
namespace BootRobust.NWDual
/-- Corollary 1, p. 13: for any distance function `R`, the robust Nadaraya–Watson budget (24)
equals the value of the perspective program over `(s, P)`. -/
theorem corollary_1 {ι : Type*} [Fintype ι] [DecidableEq ι]
(R : (ι → ℝ) → (ι → ℝ) → EReal) (D : ι → ℝ) (r : ℝ)
(w : ι → ℝ) (hw : ∀ i, 0 < w i) {Z : Type*} (L : Z → ι → ℝ) (z : Z) :
robustNW R D r w (L z) = perspValue R D r w (L z) := by sorry
end BootRobust.NWDual
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.