Lemma 2 — the nearest solution is also the nearest point of the intersection of the active half spaces
ProvedHoffmanBound.ErrorBound.lemma2linear-inequalitiesnearest-pointp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let be the solution set of the system (1), let , and let be a point of nearest to in the Euclidean distance. Let
be the half spaces of (1) whose bounding hyperplanes contain and their intersection. Then , and is a point of nearest to .
This lemma reduces the error bound to the finitely many sets of active constraints, which is where the uniform constant of the main theorem comes from.
Formalization Note The paper says "the point nearest"; the Euclidean nearest point of a nonempty closed convex set is unique, so the hypothesis is stated as " is a nearest point" and the conclusion as " is a nearest point of ". If , then .
Preamble
import Mathlib import Definitions.Def_HoffmanBound_ErrorBound_Model
Formal statement
namespace HoffmanBound.ErrorBound
open Matrix
/-- Hoffman 1952, p. 263, Lemma 2. Let `Ω` be the solution set of `Ax ≤ b`, `x ∉ Ω`, and `y` a
point of `Ω` nearest to `x` (Euclidean). Let `S` be the set of half spaces whose bounding
hyperplane contains `y` and `Ω_S` their intersection. Then `x ∉ Ω_S` and `y` is a nearest point
of `Ω_S` to `x`. -/
theorem lemma2 {m n : ℕ} (A : Matrix (Fin m) (Fin n) ℝ) (b : Fin m → ℝ) (x y : Fin n → ℝ)
(hx : x ∉ solutionSet A b) (hy : IsNearest (solutionSet A b) x y) :
x ∉ {z : Fin n → ℝ | ∀ i ∈ activeSet A b y, (A *ᵥ z) i ≤ b i} ∧
IsNearest {z : Fin n → ℝ | ∀ i ∈ activeSet A b y, (A *ᵥ z) i ≤ b i} x y := by sorry
end HoffmanBound.ErrorBound
Source
Hoffman, On Approximate Solutions of Systems of Linear Inequalities, J. Res. Nat. Bur. Standards 49 (1952), p. 263 (PDF p. 1), Lemma 2
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.