Theorem 4, p. 911, (26) (right inequality) — the infimal convolution π(r⁰, r) is at most each π^s(r⁰, r)
ProvedTractableDRO.Unified.pi_le_piOfdistributionally-robust-optimizationinfimal-convolutionp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1
Let the data be arbitrary, let be the bounds (21), (22), (24) built from them, and let . Define the unified bound by the infimal convolution
Then for every , and every ,
This is the right inequality of (26) in Theorem 4 of Goh and Sim: combining the bounds never does worse than any single one of them.
Formalization Note Values are in EReal (the printed "min" in (25) is read as an infimum), and is a Finset (Fin 3), index standing for the paper's . No hypothesis is needed for this inequality.
Preamble
import Mathlib import Definitions.Def_TractableDRO_Unified_Model open MeasureTheory ProbabilityTheory Matrix
Formal statement
namespace TractableDRO.Unified
/-- Theorem 4, p. 911, (26) (right inequality): `π(r⁰, r) ≤ π^s(r⁰, r)` for every `s ∈ S`. -/
theorem pi_le_piOf {n N Nσ : ℕ} (D : Data n N Nσ) (S : Finset (Fin 3)) (r0 : ℝ)
(r : Fin n → ℝ) : ∀ s ∈ S, piU D S r0 r ≤ piOf D s r0 r := by sorry
end TractableDRO.Unified
Source
Goh & Sim, Distributionally Robust Optimization and Its Tractable Approximations, Oper. Res. 58(4), 2010, p. 911, Theorem 4, (25), (26) (right inequality)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.