Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 3 — sup⁡{P(A):dc(μ,P)≤δ}=μ{x:c(x,A)≤1/λ∗}\sup\{P(A) : d_c(\mu,P) \le \delta\} = \mu\{x : c(x,A) \le 1/\lambda^*\}sup{P(A):dc​(μ,P)≤δ}=μ{x:c(x,A)≤1/λ∗} when c‾=c‾\underline c = \overline cc​=c

Proved
ModelRiskOT.WorstProb.theorem_3

by mikedeng1 · Oct 6, 2026 · Mathlib 0df444a (Lean v4.33.1)

distributionally-robust-optimizationoptimal-transportp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1worst-case-probability

Let SSS be a Polish space, c:S×S→R+c : S \times S \to \mathbb R_+c:S×S→R+​ a cost satisfying (A1) (nonnegative, lower semicontinuous, c(x,y)=0  ⟺  x=yc(x,y) = 0 \iff x = yc(x,y)=0⟺x=y), μ\muμ a probability measure on SSS, δ>0\delta > 0δ>0, and A⊆SA \subseteq SA⊆S a nonempty closed set. Write c(x,A)=inf⁡y∈Ac(x,y)c(x,A) = \inf_{y\in A} c(x,y)c(x,A)=infy∈A​c(x,y). Suppose that λ∗∈[0,∞)\lambda^* \in [0,\infty)λ∗∈[0,∞) attains the infimum in

inf⁡λ≥0{λδ+Eμ[(1−λc(X,A))+]}(13)\inf_{\lambda \ge 0}\Big\{\lambda\delta + E_\mu\big[(1 - \lambda c(X,A))^+\big]\Big\} \tag{13}λ≥0inf​{λδ+Eμ​[(1−λc(X,A))+]}(13)

and that the quantities c‾:=∫{c(x,A)<1/λ∗}c(x,A) dμ(x)\underline c := \int_{\{c(x,A) < 1/\lambda^*\}} c(x,A)\,d\mu(x)c​:=∫{c(x,A)<1/λ∗}​c(x,A)dμ(x) and c‾:=∫{c(x,A)≤1/λ∗}c(x,A) dμ(x)\overline c := \int_{\{c(x,A) \le 1/\lambda^*\}} c(x,A)\,d\mu(x)c:=∫{c(x,A)≤1/λ∗}​c(x,A)dμ(x) of (14) are equal. Then

sup⁡{P(A):dc(μ,P)≤δ}=μ{x:c(x,A)≤1/λ∗}.(15)\sup\{P(A) : d_c(\mu,P) \le \delta\} = \mu\{x : c(x,A) \le 1/\lambda^*\}. \tag{15}sup{P(A):dc​(μ,P)≤δ}=μ{x:c(x,A)≤1/λ∗}.(15)

The worst-case probability of AAA over all models within optimal transport cost δ\deltaδ of the baseline is the baseline probability of the inflated set {x:c(x,A)≤1/λ∗}\{x : c(x,A) \le 1/\lambda^*\}{x:c(x,A)≤1/λ∗}. No optimal transport plan is assumed to exist.

Formalization Note The threshold c(x,A)≤1/λ∗c(x,A) \le 1/\lambda^*c(x,A)≤1/λ∗ is written λ∗c(x,A)≤1\lambda^* c(x,A) \le 1λ∗c(x,A)≤1, which is the whole space at λ∗=0\lambda^* = 0λ∗=0 (the paper's 1/0=∞1/0 = \infty1/0=∞; then both sides equal 111). The supremum is over probability measures PPP on SSS with dc(μ,P)≤δd_c(\mu,P) \le \deltadc​(μ,P)≤δ, with both marginals of the couplings in dcd_cdc​ fixed. Both sides are in [0,∞][0,\infty][0,∞]; μ\muμ of the inflated set, which is universally measurable, is its outer measure, i.e. its completion measure.

Preamble
import Mathlib
import Definitions.Def_ModelRiskOT_WorstProb_WorstCaseProb

open MeasureTheory
open scoped ENNReal
Formal statement
namespace ModelRiskOT.WorstProb

/-- Theorem 3, p. 10, (15). Under (A1), for a nonempty closed `A`, if `λ* ∈ [0, ∞)` attains the
infimum in (13) and `c̲ = c̄` (14), then
`sup {P(A) : d_c(μ, P) ≤ δ} = μ {x : c(x, A) ≤ 1/λ*}`.
The threshold is in multiplied form `λ* c(x, A) ≤ 1` (the whole space at `λ* = 0`, i.e. `1/0 = ∞`).

Sanity check (not part of the statement): `S = ℝ`, `c(x, y) = |x − y|`, `μ` uniform on `[0, 1]`,
`A = [1, ∞)`, `0 < δ ≤ 1/2`. Then `c(x, A) = 1 − x` on `[0, 1]`, `g(λ) = λδ + 1/(2λ)` for `λ ≥ 1`,
minimized at `λ* = 1/√(2δ)` with value `√(2δ)`, `c̲ = c̄ = δ`, and
`μ {x : 1 − x ≤ √(2δ)} = √(2δ)`: both sides agree and are strictly between 0 and 1.
With `μ` the point mass at `0` and `δ < 1`, instead, `λ* = 1`, `c̲ = 0 ≠ 1 = c̄`: the hypothesis
fails, and indeed `I = δ ≠ 1`. -/
theorem theorem_3 {S : Type*} [TopologicalSpace S] [PolishSpace S] [MeasurableSpace S]
    [BorelSpace S] (c : S → S → ℝ) (hc : ModelRiskOT.Duality.AssumptionA1 c) (μ : Measure S) [IsProbabilityMeasure μ]
    (δ : ℝ) (hδ : 0 < δ) (A : Set S) (hA : IsClosed A) (hAne : A.Nonempty)
    (lamStar : ℝ) (hlam : AttainsInf13 c μ δ A lamStar)
    (hcc : cLower c μ A lamStar = cUpper c μ A lamStar) :
    worstProb c μ δ A = μ {x | lamStar * costToSet c A x ≤ 1} := by sorry

end ModelRiskOT.WorstProb
Source
Blanchet & Murthy, Quantifying Distributional Model Risk via Optimal Transport, arXiv:1604.01446v2, p. 10, Theorem 3, Eq. (15)
Human review
  • Endorsed by Shuze Chen · Oct 6, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 6, 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