Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proposition 2 (a) and λ-part of (b), (12c)–(12d), p. 16 — one-sided λ-derivatives of ℓ_rob are min/max of −√δ(γ²βᵀA(x)⁻¹β − 1) over Γ*

Proved
OTDRO.WorstCase.ellRob_lambda_derivatives

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

directional-derivativedistributionally-robust-optimizationenvelope-theoremp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1

Assume Assumption 1, δ>0\delta>0δ>0, B⊆RdB\subseteq\mathbb R^dB⊆Rd convex, and let ℓ\ellℓ satisfy Assumption 2 and be of the form ℓ(u)=max⁡i=1,…,Kℓi(u)\ell(u)=\max_{i=1,\dots,K}\ell_i(u)ℓ(u)=maxi=1,…,K​ℓi​(u) for a positive integer KKK and continuously differentiable ℓi:R→R\ell_i:\mathbb R\to\mathbb Rℓi​:R→R. Let U1={(β,λ):β∈B, λ>λthr(β)}\mathbb U_1=\{(\beta,\lambda):\beta\in B,\ \lambda>\lambda_{thr}(\beta)\}U1​={(β,λ):β∈B, λ>λthr​(β)} and, for γ∈R\gamma\in\mathbb Rγ∈R,

∂λF(γ,β,λ;x)=−δ(γ2βTA(x)−1β−1).\partial_\lambda F(\gamma,\beta,\lambda;x)=-\sqrt\delta\bigl(\gamma^2\beta^{\mathsf T}A(x)^{-1}\beta-1\bigr).∂λ​F(γ,β,λ;x)=−δ​(γ2βTA(x)−1β−1).

Then for P0P_0P0​-almost every xxx, for every (β,λ)∈U1(\beta,\lambda)\in\mathbb U_1(β,λ)∈U1​:

  1. (a) Γ∗(β,λ;x)≠∅\Gamma^*(\beta,\lambda;x)\neq\emptysetΓ∗(β,λ;x)=∅;
  2. λ↦ℓrob(β,λ;x)\lambda\mapsto\ell_{\rm rob}(\beta,\lambda;x)λ↦ℓrob​(β,λ;x) is finite near λ\lambdaλ and absolutely continuous on every interval with endpoints in (λthr(β),∞)(\lambda_{thr}(\beta),\infty)(λthr​(β),∞);
  3. (12c), (12d): its left and right derivatives are
∂−ℓrob∂λ(β,λ;x)=min⁡γ∈Γ∗(β,λ;x)−δ(γ2βTA(x)−1β−1),∂+ℓrob∂λ(β,λ;x)=max⁡γ∈Γ∗(β,λ;x)−δ(γ2βTA(x)−1β−1),\frac{\partial_-\ell_{\rm rob}}{\partial\lambda}(\beta,\lambda;x)=\min_{\gamma\in\Gamma^*(\beta,\lambda;x)}-\sqrt\delta\bigl(\gamma^2\beta^{\mathsf T}A(x)^{-1}\beta-1\bigr),\qquad \frac{\partial_+\ell_{\rm rob}}{\partial\lambda}(\beta,\lambda;x)=\max_{\gamma\in\Gamma^*(\beta,\lambda;x)}-\sqrt\delta\bigl(\gamma^2\beta^{\mathsf T}A(x)^{-1}\beta-1\bigr),∂λ∂−​ℓrob​​(β,λ;x)=γ∈Γ∗(β,λ;x)min​−δ​(γ2βTA(x)−1β−1),∂λ∂+​ℓrob​​(β,λ;x)=γ∈Γ∗(β,λ;x)max​−δ​(γ2βTA(x)−1β−1),

both minimum and maximum being attained; 4. λ↦ℓrob(β,λ;x)\lambda\mapsto\ell_{\rm rob}(\beta,\lambda;x)λ↦ℓrob​(β,λ;x) is differentiable at λ\lambdaλ if and only if {∂λF(γ,β,λ;x):γ∈Γ∗(β,λ;x)}\{\partial_\lambda F(\gamma,\beta,\lambda;x):\gamma\in\Gamma^*(\beta,\lambda;x)\}{∂λ​F(γ,β,λ;x):γ∈Γ∗(β,λ;x)} is a singleton.

This is the envelope-theorem description of the robust loss in the multiplier λ\lambdaλ; integrated over xxx it yields the one-sided derivatives of the dual objective used to locate the dual optimizer.

Formalization Note Only part (a) and the λ\lambdaλ-part of part (b) are formalized; the βj\beta_jβj​-derivatives (12a), (12b), the absolute continuity in βj\beta_jβj​, the βj\beta_jβj​-differentiability criterion and (13) are omitted. Because FFF is affine in λ\lambdaλ, both one-sided partial derivatives ∂±F/∂λ\partial_\pm F/\partial\lambda∂±​F/∂λ equal −δ(γ2βTA(x)−1β−1)-\sqrt\delta(\gamma^2\beta^{\mathsf T}A(x)^{-1}\beta-1)−δ​(γ2βTA(x)−1β−1), so the page's set {∂+F/∂λ,∂−F/∂λ:γ∈Γ∗}\{\partial_+F/\partial\lambda,\partial_-F/\partial\lambda:\gamma\in\Gamma^*\}{∂+​F/∂λ,∂−​F/∂λ:γ∈Γ∗} is the image written here. One-sided derivatives are HasDerivWithinAt on (−∞,λ](-\infty,\lambda](−∞,λ] and [λ,∞)[\lambda,\infty)[λ,∞) of the real-valued map l↦ℓrob(β,l;x)l\mapsto\ell_{\rm rob}(\beta,l;x)l↦ℓrob​(β,l;x), which is finite near λ\lambdaλ (stated as a conjunct). "Absolutely continuous for (β,λ)∈U1(\beta,\lambda)\in\mathbb U_1(β,λ)∈U1​" is read as absolute continuity on every compact interval inside (λthr(β),∞)(\lambda_{thr}(\beta),\infty)(λthr​(β),∞). The "P0P_0P0​-almost every xxx" quantifier is placed before (β,λ)(\beta,\lambda)(β,λ), as on the page: one null set serves all (β,λ)∈U1(\beta,\lambda)\in\mathbb U_1(β,λ)∈U1​. ℓ=max⁡iℓi\ell=\max_i\ell_iℓ=maxi​ℓi​ is written as: ℓ(u)\ell(u)ℓ(u) is the greatest element of {ℓi(u)}\{\ell_i(u)\}{ℓi​(u)}.

Preamble
import Mathlib
import Definitions.Def_OTDRO_Dual_Setting

open MeasureTheory Filter Topology
Formal statement
namespace OTDRO.WorstCase

/-- **Proposition 2**, part (a) and the `λ`-part of (b) (arXiv:1810.02403v3, p. 16; proof
pp. 48–49): for `ℓ = max_{i ≤ K} ℓᵢ` with `ℓᵢ` continuously differentiable and `ℓ` satisfying
Assumption 2, for `P₀`-almost every `x` and every `(β, λ) ∈ 𝕌₁`: `Γ*(β, λ; x) ≠ ∅`;
`λ ↦ ℓ_rob(β, λ; x)` is finite near `λ` and absolutely continuous on every interval inside
`(λ_thr(β), ∞)`; its left and right derivatives are the minimum (12c) and the maximum (12d) of
`−√δ(γ² βᵀA(x)⁻¹β − 1)` over `γ ∈ Γ*(β, λ; x)`; and it is differentiable at `λ` if and only if
the set `{∂±F/∂λ(γ, β, λ; x) : γ ∈ Γ*(β, λ; x)} = {−√δ(γ²βᵀA(x)⁻¹β − 1) : γ ∈ Γ*}` is a
singleton. -/
theorem ellRob_lambda_derivatives {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 : OTDRO.Dual.Assumption1 P0 A ρmin ρmax) (ℓ : ℝ → ℝ) (h2 : OTDRO.Dual.Assumption2 P0 ℓ)
    (δ : ℝ) (hδ : 0 < δ) (B : Set (EuclideanSpace ℝ (Fin d))) (hB : Convex ℝ B)
    (K : ℕ) (hK : 0 < K) (ℓi : Fin K → ℝ → ℝ) (hℓi : ∀ i, ContDiff ℝ 1 (ℓi i))
    (hmax : ∀ u, IsGreatest (Set.range fun i => ℓi i u) (ℓ u)) :
    ∀ᵐ x ∂P0, ∀ β ∈ B, ∀ lam : ℝ, OTDRO.Dual.lamThr P0 ℓ A δ β < lam →
      (OTDRO.Dual.maximizers ℓ A δ β lam x).Nonempty ∧
      (∀ᶠ l in 𝓝 lam, OTDRO.Dual.ellRob ℓ A δ β l x ≠ ⊤) ∧
      (∀ a b : ℝ, OTDRO.Dual.lamThr P0 ℓ A δ β < a → OTDRO.Dual.lamThr P0 ℓ A δ β < b →
        AbsolutelyContinuousOnInterval (fun l => (OTDRO.Dual.ellRob ℓ A δ β l x).toReal) a b) ∧
      (∃ m : ℝ, IsLeast ((fun γ => -Real.sqrt δ * (γ ^ 2 * OTDRO.Dual.quadInv A β x - 1)) ''
          OTDRO.Dual.maximizers ℓ A δ β lam x) m ∧
        HasDerivWithinAt (fun l => (OTDRO.Dual.ellRob ℓ A δ β l x).toReal) m (Set.Iic lam) lam) ∧
      (∃ M : ℝ, IsGreatest ((fun γ => -Real.sqrt δ * (γ ^ 2 * OTDRO.Dual.quadInv A β x - 1)) ''
          OTDRO.Dual.maximizers ℓ A δ β lam x) M ∧
        HasDerivWithinAt (fun l => (OTDRO.Dual.ellRob ℓ A δ β l x).toReal) M (Set.Ici lam) lam) ∧
      (DifferentiableAt ℝ (fun l => (OTDRO.Dual.ellRob ℓ A δ β l x).toReal) lam ↔
        ∃ s : ℝ, (fun γ => -Real.sqrt δ * (γ ^ 2 * OTDRO.Dual.quadInv A β x - 1)) ''
          OTDRO.Dual.maximizers ℓ A δ β lam x = {s}) := by sorry

end OTDRO.WorstCase
Source
arXiv:1810.02403v3, Proposition 2(a) and (b) (12c)–(12d), p. 16; proof pp. 48–49
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