Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Section 2, equations (3)–(4) — no duality gap

Disproved
WangMinimaxRegret.Reformulation.eq_3_4

by Shuze Chen · Oct 29, 2026 · Mathlib 0df444a (Lean v4.33.1)

linear-optimizationmechanism-designoperations-researchp2o-gran-per-chapterp2o-plan-paperp2o-v1

Fix consistent nonnegative valuation data with arbitrary K≥0K\ge0K≥0, a feasible nondecreasing right-continuous normalized Stieltjes payment mmm, and a finite price p∈Ip\in Ip∈I. Let V3V_3V3​ be the infimum of ∫m dH\int m\,dH∫mdH over finite nonnegative measures on III satisfying the finite centered moment equations and H([p,∞))=1H([p,\infty))=1H([p,∞))=1. Let D4D_4D4​ be the supremum of real θ\thetaθ for which real multipliers ℓ1,…,ℓK\ell_1,\ldots,\ell_Kℓ1​,…,ℓK​ satisfy

∑k=1K(xk−μk)ℓk≤m(x) (x∈I, x<p),θ+∑k=1K(xk−μk)ℓk≤m(x) (x∈I, x≥p).\sum_{k=1}^K(x^k-\mu_k)\ell_k\le m(x)\ (x\in I,\ x<p),\qquad \theta+\sum_{k=1}^K(x^k-\mu_k)\ell_k\le m(x)\ (x\in I,\ x\ge p).k=1∑K​(xk−μk​)ℓk​≤m(x) (x∈I, x<p),θ+k=1∑K​(xk−μk​)ℓk​≤m(x) (x∈I, x≥p).

The source assertion, retained here as printed, is

V3=D4,V_3=D_4,V3​=D4​,

and whenever V3V_3V3​ equals a finite real vvv, there exist multipliers making (v,ℓ)(v,\ell)(v,ℓ) dual feasible. This includes the finite-value attainment indicated by the displayed maximum. It assumes neither Slater regularity, positive variance, continuity of mmm, nor an existing dual optimizer. An infeasible primal has extended value +∞+\infty+∞, with no assertion of an infinite real maximizing variable. The claim is the duality step used before Theorem 1; the source counterexample is documented separately in INSIGHTS.yaml, not excluded by an added hypothesis.

Differences from the source:

  • encoding: real interval, payment functions, probability laws, and moment orders 1 through K → real-line Borel measures supported on the interval, Stieltjes functions extended by zero below a and constantly above finite b, and Fin K exponents i+1; reason: extensions only fix values outside the domain; all atoms and payment jumps are retained, including the payment atom at a>0
  • encoding: the seller knows moments of an actual valuation distribution → Consistent d requires a nonempty printed ambiguity class, with finite specified moments; reason: makes the standing information-consistency assumption explicit; no interiority, positive variance, or density is assumed
  • encoding: nonnegative revenue integrals and optimization infima/suprema → nonnegative extended integrals and ENNReal/EReal extrema; zero-tail laws impose no fractional constraint; reason: avoids nonintegrable real integrals defaulting to zero and empty real extrema; empty infima are positive infinity
  • encoding: real homogeneous moment integrals in (3) → finite H and integrable centered moment tests, omitting the identically zero order-zero row; reason: preserves the well-defined scalar rescaling domain of (3)
  • reading: no duality gap and a displayed max in (4) → extended-value equality and dual attainment whenever the primal value is a finite real; reason: makes both the equality and maximum clauses explicit without assuming attainment
  • reading: upper support endpoint b, used as infinity in later applications → b may be infinite, with all quantified valuations and prices finite; reason: no compactness or bounded-total-payment condition is added
Preamble
import Definitions.Def_WangMinimaxRegret_Reformulation_Programs

open MeasureTheory Set
open scoped ENNReal NNReal
set_option autoImplicit false
Formal statement
namespace WangMinimaxRegret.Reformulation

theorem eq_3_4 (d : Data) (hc : Consistent d) (m : StieltjesFunction ℝ)
    (hm : IsPayment d m) (p : ℝ) (hp : p ∈ interval d) :
    ((homogenizedLP d m p).value : EReal) = dualValue d m p ∧
    (∀ v : ℝ, ((homogenizedLP d m p).value : EReal) = (v : EReal) →
      ∃ lam : Fin d.K → ℝ, DualFeasible d m p v lam) := by sorry

end WangMinimaxRegret.Reformulation
Source
Wang, Liu and Zhang (2024), Minimax Regret Robust Screening with Moment Information, https://doi.org/10.1287/msom.2023.0072, Section 2, p. 9, equation (4) and no-gap assertion referring to (3)
Read-back

What the Lean code literally says, in plain math · claude-opus-5-5

Setting. Fix a data record ddd:

  • reals a≥0a\ge 0a≥0 and b∈R∪{+∞}b\in\mathbb R\cup\{+\infty\}b∈R∪{+∞} with a≤ba\le ba≤b;
  • K∈NK\in\mathbb NK∈N and reals μ0,…,μK−1\mu_0,\dots,\mu_{K-1}μ0​,…,μK−1​;
  • the interval I={x:a≤x≤b}I=\{x : a\le x\le b\}I={x:a≤x≤b}.

Hypotheses.

  1. ddd is consistent: there is a Borel probability measure FFF on R\mathbb RR such that:
    • F(R∖I)=0F(\mathbb R\setminus I)=0F(R∖I)=0;
    • xi+1x^{i+1}xi+1 is FFF-integrable with ∫xi+1 dF=μi\int x^{i+1}\,dF = \mu_i∫xi+1dF=μi​ for all i<Ki<Ki<K;
    • F((−∞,a])=0F((-\infty,a])=0F((−∞,a])=0.
  2. mmm is a Stieltjes function (nondecreasing, right-continuous, real-valued) that is a payment for ddd:
    • m(0)=0m(0)=0m(0)=0;
    • m=0m=0m=0 on (−∞,a)(-\infty,a)(−∞,a);
    • mmm is constant on {x≥b}\{x\ge b\}{x≥b};
    • m≥0m\ge0m≥0 on III;
    • ∫I∩(0,x]t−1 dνm(t)≤1\int_{I\cap(0,x]} t^{-1}\,d\nu_m(t)\le 1∫I∩(0,x]​t−1dνm​(t)≤1 for x∈Ix\in Ix∈I, with νm\nu_mνm​ the Lebesgue–Stieltjes measure of mmm.
  3. p∈Ip\in Ip∈I.

Quantities used.

  • Primal value. V∈[0,∞]V\in[0,\infty]V∈[0,∞] is the infimum of ∫max⁡(m(x),0) dH(x)\int\max(m(x),0)\,dH(x)∫max(m(x),0)dH(x) over Borel measures HHH on R\mathbb RR such that:

    • HHH is finite;
    • H(R∖I)=0H(\mathbb R\setminus I)=0H(R∖I)=0;
    • H([p,∞))=1H([p,\infty))=1H([p,∞))=1;
    • for each j<Kj<Kj<K, the function xj+1−μjx^{j+1}-\mu_jxj+1−μj​ is HHH-integrable and ∫(xj+1−μj) dH=0\int(x^{j+1}-\mu_j)\,dH = 0∫(xj+1−μj​)dH=0.

    It is +∞+\infty+∞ if no such HHH exists.

  • Moment polynomial. For λ∈RK\lambda\in\mathbb R^Kλ∈RK, Pλ(x)=∑i<K(xi+1−μi)λiP_\lambda(x)=\sum_{i<K}(x^{i+1}-\mu_i)\lambda_iPλ​(x)=∑i<K​(xi+1−μi​)λi​.

  • Dual feasibility. A pair (θ,λ)∈R×RK(\theta,\lambda)\in\mathbb R\times\mathbb R^K(θ,λ)∈R×RK is dual-feasible when:

    • Pλ(x)≤m(x)P_\lambda(x)\le m(x)Pλ​(x)≤m(x) for all x∈Ix\in Ix∈I with x<px<px<p;
    • θ+Pλ(x)≤m(x)\theta+P_\lambda(x)\le m(x)θ+Pλ​(x)≤m(x) for all x∈Ix\in Ix∈I with x≥px\ge px≥p.
  • Dual value. D∈R‾D\in\overline{\mathbb R}D∈R is the supremum of θ\thetaθ over dual-feasible pairs.

Conclusions. Both of the following hold.

  1. Equal values. V=DV = DV=D as elements of [−∞,+∞][-\infty,+\infty][−∞,+∞].
  2. Dual attainment. For every real vvv, if V=vV = vV=v (that is, VVV is finite and equals vvv), then there exists λ∈RK\lambda\in\mathbb R^Kλ∈RK such that (v,λ)(v,\lambda)(v,λ) is dual-feasible.

Degenerate cases.

  • Sign of the dual value. V≥0V\ge 0V≥0 always, so conclusion 1 entails in particular that D≠−∞D \ne -\inftyD=−∞.
  • Infinite primal value. If V=+∞V=+\inftyV=+∞, conclusion 1 asserts that dual-feasible θ\thetaθ are unbounded above, and conclusion 2 is vacuous.
  • K=0K=0K=0.
    • λ\lambdaλ ranges over the single empty vector and Pλ≡0P_\lambda\equiv 0Pλ​≡0.
    • Dual feasibility becomes 0≤m0\le m0≤m on I∩(−∞,p)I\cap(-\infty,p)I∩(−∞,p) and θ≤m\theta\le mθ≤m on I∩[p,∞)I\cap[p,\infty)I∩[p,∞).
    • The primal requires only a finite measure on III with unit mass on [p,∞)[p,\infty)[p,∞).
  • p=ap = ap=a. The first dual condition is vacuous.
  • p=bp = bp=b (finite). The second dual condition concerns only x=bx = bx=b.
  • No normalization of total mass. Nothing fixes H(R)H(\mathbb R)H(R) in the primal.
Human review
  • Endorsed by Community (Bot) · Oct 29, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Shuze Chen · Oct 29, 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