Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 1, p. 10 — Mahalanobis transport dual reformulation and attainment

Proved
OTDRO.Dual.dual_reformulation

by mikedeng1 · Oct 7, 2026 · Mathlib 0df444a (Lean v4.33.1)

distributionally-robust-optimizationdualityoptimal-transportp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1

Let P0P_0P0​ be a Borel probability law on Rd\mathbb R^dRd. Let AAA satisfy Assumption 1, let δ>0\delta>0δ>0, and let B⊆RdB\subseteq\mathbb R^dB⊆Rd be convex. For a decision β∈B\beta\in Bβ∈B, suppose ℓ:R→R\ell:\mathbb R\to\mathbb Rℓ:R→R is upper semicontinuous and x↦ℓ(βTx)x\mapsto\ell(\beta^{\mathsf T}x)x↦ℓ(βTx) is integrable under P0P_0P0​. The worst case over transport plans of cost at most δ\deltaδ satisfies

sup⁡P: Dc(P0,P)≤δEP[ℓ(βTX)]=inf⁡λ≥0fδ(β,λ).\sup_{P:\,D_c(P_0,P)\le\delta}E_P[\ell(\beta^{\mathsf T}X)]=\inf_{\lambda\ge0}f_\delta(\beta,\lambda).P:Dc​(P0​,P)≤δsup​EP​[ℓ(βTX)]=λ≥0inf​fδ​(β,λ).

Moreover, some λ∗≥0\lambda^*\ge0λ∗≥0 attains the infimum: fδ(β,λ∗)=inf⁡λ≥0fδ(β,λ)f_\delta(\beta,\lambda^*)=\inf_{\lambda\ge0}f_\delta(\beta,\lambda)fδ​(β,λ∗)=infλ≥0​fδ​(β,λ). The reformulation expresses the worst-case loss through an expectation under the baseline law and a one dimensional dual search.

Formalization Note The left side uses the paper's equivalent coupling linear program from §2.1. Both sides live in the extended reals. Integrability of ℓ(βTX)\ell(\beta^{\mathsf T}X)ℓ(βTX) is stated explicitly because the cited strong duality theorem [6, Theorem 1] requires it; Theorem 1 on p. 10 omits this condition. The standing convexity of BBB and β∈B\beta\in Bβ∈B are retained.

Preamble
import Mathlib
import Definitions.Def_OTDRO_Dual_Setting
import Definitions.Def_ModelRiskOT_Duality_primalValue
Formal statement
namespace OTDRO.Dual

open MeasureTheory

/-- Theorem 1, p. 10: the Mahalanobis-transport worst-case loss has the one-dimensional
dual representation, and the dual infimum is attained for decisions in `B`. -/
theorem dual_reformulation {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 : Assumption1 P0 A ρmin ρmax)
    (ℓ : ℝ → ℝ) (hℓ : UpperSemicontinuous ℓ)
    (δ : ℝ) (hδ : 0 < δ)
    (B : Set (EuclideanSpace ℝ (Fin d))) (hB : Convex ℝ B)
    (β : EuclideanSpace ℝ (Fin d)) (hβ : β ∈ B)
    (hint : Integrable (fun x => ℓ (inner ℝ β x)) P0) :
    ModelRiskOT.Duality.primalValue (mahalCost A) (fun x => ℓ (inner ℝ β x)) P0 δ =
        ⨅ lam ∈ Set.Ici (0 : ℝ), fDelta P0 ℓ A δ β lam ∧
      ∃ lamStar : ℝ, 0 ≤ lamStar ∧
        fDelta P0 ℓ A δ β lamStar =
          ⨅ lam ∈ Set.Ici (0 : ℝ), fDelta P0 ℓ A δ β lam := by sorry

end OTDRO.Dual
Source
arXiv:1810.02403v3, Theorem 1, p. 10; §5.1 proof, p. 32
Human review
  • Endorsed by Shuze Chen · Oct 8, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 8, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me