Proof of Theorem 5′, second display — as sets of perturbed matrices, hence
ProvedRobustRegLasso.Sparsity.theorem_5_prime_inclusionLet , radii , numbers , and an index set , with for and for . Then
and consequently, for every ,
This is the second step of the proof of Theorem 5′: enlarging the bounds on the features outside by absorbs the change from to , so the worst case for dominates the worst case for at every point.
Formalization Note The printed display reads , with and exchanged relative to the inclusion that justifies it and to the final display of the proof that uses it. The version stated here is the one the inclusion gives and the proof uses (the printed one is also true, by the symmetric inclusion). Matrices are column families, so is the family . Maxima are suprema in EReal. No sign condition on or is needed.
import Mathlib import Definitions.Def_RobustRegLasso_Sparsity_Basic
namespace RobustRegLasso.Sparsity
theorem theorem_5_prime_inclusion {n m : ℕ} (a aTilde : Fin m → EuclideanSpace ℝ (Fin n))
(b : EuclideanSpace ℝ (Fin n)) (c l : Fin m → ℝ) (I : Finset (Fin m))
(hl : ∀ j, j ∉ I → ‖a j - aTilde j‖ ≤ l j) (hI : ∀ i, i ∈ I → a i = aTilde i) :
{M | ∃ δ ∈ (RobustRegLasso.FeatureWise.uncertaintySet c : Set (Fin m → EuclideanSpace ℝ (Fin n))), M = a + δ} ⊆
{M | ∃ δ ∈ (enlargedSet c l I : Set (Fin m → EuclideanSpace ℝ (Fin n))), M = aTilde + δ} ∧
∀ x' : Fin m → ℝ, robustObjective a b (RobustRegLasso.FeatureWise.uncertaintySet c) x' ≤
robustObjective aTilde b (enlargedSet c l I) x' := by sorry
end RobustRegLasso.Sparsity
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.