Lemma 3(a), p. 32 — every g ∈ Γ* satisfies √δ|g|βᵀA(x)⁻¹β ≤ 1 + C₁ε⁻¹(1 + |βᵀx|) once λ ≥ (κ + ε)√δβᵀA(x)⁻¹β
ProvedOTDRO.WorstCase.maximizers_boundedcompactnessdistributionally-robust-optimizationp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Assume Assumptions 1 and 2, and convex, and let be the growth exponent of . For every , , and , there is a constant such that every maximizer of satisfies
The bound shows that the maximizer set is bounded, hence compact, as soon as exceeds the threshold, and that the worst-case displacement grows at most linearly in . It is what makes the extreme points and of over attained in the proof of Theorem 6(c), and what makes the moments finite.
Formalization Note The constant is quantified after , , and , as in the printed lemma. Only part (a) of Lemma 3 is formalized here; part (b) is in the sibling mission on convexity of the dual objective.
Preamble
import Mathlib import Definitions.Def_OTDRO_Dual_Setting open MeasureTheory
Formal statement
namespace OTDRO.WorstCase
/-- **Lemma 3(a)** (arXiv:1810.02403v3, p. 32; proof p. 47), with the constant `C₁` chosen after
`ε`, `x`, `β` and `λ`, as the printed lemma states: if
`λ ≥ (κ + ε)√δ βᵀA(x)⁻¹β`, every `g ∈ Γ*(β, λ; x)` satisfies
`√δ |g| βᵀA(x)⁻¹β ≤ 1 + C₁ ε⁻¹ (1 + |βᵀx|)`. -/
theorem maximizers_bounded {d : ℕ} (P0 : Measure (EuclideanSpace ℝ (Fin d)))
[IsProbabilityMeasure P0] (A : EuclideanSpace ℝ (Fin d) → Matrix (Fin d) (Fin d) ℝ)
(ρmin ρmax : ℝ) (hA : OTDRO.Dual.Assumption1 P0 A ρmin ρmax) (ℓ : ℝ → ℝ) (h2 : OTDRO.Dual.Assumption2 P0 ℓ)
(δ : ℝ) (hδ : 0 < δ) (B : Set (EuclideanSpace ℝ (Fin d))) (hB : Convex ℝ B) :
∀ ε : ℝ, 0 < ε → ∀ x : EuclideanSpace ℝ (Fin d), ∀ β ∈ B, ∀ lam : ℝ,
(OTDRO.Dual.growthKappa ℓ + ε) * Real.sqrt δ * OTDRO.Dual.quadInv A β x ≤ lam →
∃ C1 : ℝ, 0 < C1 ∧
∀ g ∈ OTDRO.Dual.maximizers ℓ A δ β lam x,
Real.sqrt δ * |g| * OTDRO.Dual.quadInv A β x ≤ 1 + C1 * ε⁻¹ * (1 + |inner ℝ β x|) := by sorry
end OTDRO.WorstCase
Source
arXiv:1810.02403v3, Lemma 3(a), p. 32; proof p. 47
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.