Corollary 4.3, p. 14 — (10) ≤ optimal value of (12f) for any ε ≥ 0, without Assumption 4.1
ProvedWassDDRO.Reduction.corollary_4_3convex-conjugatedistributionally-robust-optimizationp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1upper-boundwasserstein
Let be a finite-dimensional real normed space with its Borel σ-algebra and dual norm , , measurable () with , samples () and . Then the worst-case expectation (10) is at most the optimal value of the finite convex program (12f):
No convexity of or of the is assumed.
This gives a computable conservative bound on the worst-case expectation for arbitrary piecewise losses.
Formalization Note Here , which is the paper's reading of the sum under (no EReal addition is performed). The infimum is when (12f) is infeasible.
Preamble
import Mathlib import Definitions.Def_WassDDRO_Reduction_Setting open MeasureTheory Filter Topology
Formal statement
namespace WassDDRO.Reduction
/-- Corollary 4.3 (Approximate convex reduction), p. 14: for any `ε ≥ 0`, without
Assumption 4.1, the worst-case expectation (10) is at most the optimal value of (12f). -/
theorem corollary_4_3 {E : Type*} [NormedAddCommGroup E] [NormedSpace ℝ E] [FiniteDimensional ℝ E]
[MeasurableSpace E] [BorelSpace E]
{K N : ℕ} (hK : 0 < K) (hN : 0 < N) (Ξ : Set E) (ℓ : Fin K → E → EReal)
(hmeas : ∀ k, Measurable (ℓ k)) (ξhat : Fin N → E) (hξ : ∀ i, ξhat i ∈ Ξ)
(ε : ℝ) (hε : 0 ≤ ε) :
worstCaseExpectation ε Ξ ξhat (maxLoss ℓ) ≤ program12fValue ε Ξ ξhat ℓ := by sorry
end WassDDRO.Reduction
Source
Mohajerin Esfahani & Kuhn, arXiv:1505.05116v3, Corollary 4.3, p. 14; program (12f), p. 13
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.