Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 2.1, p. 4 — |v(ξ) − v(ξ̃)| ≤ L(‖ξ − ξ̃‖_r + Σ_{t=2}^{T−1} D_t(F_t, F̃_t))

Proved
MultistageStab.Value.theorem_2_1

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

filtration-distancemultistagep2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1stabilitystochastic-programming

Consider the linear multistage stochastic program (1) with input process ξ=(ξ1,…,ξT)\xi = (\xi_1,\dots,\xi_T)ξ=(ξ1​,…,ξT​), polyhedral X1X_1X1​, polyhedral cones XtX_tXt​ (t≥2t \ge 2t≥2), fixed recourse matrices, and costs btb_tbt​ and right-hand sides hth_tht​ affine in ξt\xi_tξt​, under one of the three randomness patterns with exponents (r,r′)(r,r')(r,r′). Assume

  1. (A1) complete fixed recourse: At,0Xt=RntA_{t,0}X_t = \mathbb R^{n_t}At,0​Xt​=Rnt​ for t=2,…,Tt = 2,\dots,Tt=2,…,T;
  2. (A2) v(ξ)v(\xi)v(ξ) is finite and the objective FFF is level-bounded locally uniformly at ξ\xiξ;
  3. (A3) ξ∈Lr(Ω,F,P;RTd)\xi \in L_r(\Omega,\mathcal F,\mathbb P;\mathbb R^{Td})ξ∈Lr​(Ω,F,P;RTd), with ξ1\xi_1ξ1​ deterministic;
  4. X1X_1X1​ is bounded.

Then there are positive constants LLL, α\alphaα, δ\deltaδ such that

∣v(ξ)−v(ξ~)∣≤L(∥ξ−ξ~∥r+∑t=2T−1Dt(Ft,F~t))|v(\xi) - v(\tilde\xi)| \le L\Big( \|\xi - \tilde\xi\|_r + \sum_{t=2}^{T-1} D_t(\mathcal F_t, \tilde{\mathcal F}_t) \Big)∣v(ξ)−v(ξ~​)∣≤L(∥ξ−ξ~​∥r​+t=2∑T−1​Dt​(Ft​,F~t​))

for every admissible input ξ~\tilde\xiξ~​ with ∥ξ~−ξ∥r≤δ\|\tilde\xi - \xi\|_r \le \delta∥ξ~​−ξ∥r​≤δ and v(ξ~)v(\tilde\xi)v(ξ~​) finite. Here Ft\mathcal F_tFt​ and F~t\tilde{\mathcal F}_tF~t​ are the filtrations generated by ξ\xiξ and ξ~\tilde\xiξ~​, and

Dt(Ft,F~t)=max⁡{sup⁡x∈lα(F(ξ,⋅))∥xt−E[xt∣F~t]∥r′, sup⁡x~∈lα(F(ξ~,⋅))∥x~t−E[x~t∣Ft]∥r′},D_t(\mathcal F_t,\tilde{\mathcal F}_t) = \max\Big\{ \sup_{x \in l_\alpha(F(\xi,\cdot))} \|x_t - \mathbb E[x_t \mid \tilde{\mathcal F}_t]\|_{r'},\ \sup_{\tilde x \in l_\alpha(F(\tilde\xi,\cdot))} \|\tilde x_t - \mathbb E[\tilde x_t \mid \mathcal F_t]\|_{r'} \Big\},Dt​(Ft​,F~t​)=max{x∈lα​(F(ξ,⋅))sup​∥xt​−E[xt​∣F~t​]∥r′​, x~∈lα​(F(ξ~​,⋅))sup​∥x~t​−E[x~t​∣Ft​]∥r′​},

both level sets taken at threshold v(ξ)+αv(\xi) + \alphav(ξ)+α.

The theorem shows that the optimal value of a multistage program depends on the input not only through an LrL_rLr​ distance but also through a distance between the information structures, and that the two together control it in a Lipschitz way.

Formalization Note The estimate is stated in [0,∞][0,\infty][0,∞] (ENNReal.ofReal |v(ξ) − v(ξ̃)| ≤ …); both optimal values are finite extended reals, so their real parts are the values themselves. The randomness pattern is a hypothesis on the data, and all three rows of the paper's table are covered. Admissible inputs are measurable, in LrL_rLr​, and have a deterministic first component, as the model's F1={∅,Ω}\mathcal F_1 = \{\emptyset,\Omega\}F1​={∅,Ω} requires.

Preamble
import Mathlib
import Definitions.Def_MultistageStab_Value_Setting

open MeasureTheory ProbabilityTheory
open scoped ENNReal
Formal statement
namespace MultistageStab.Value

theorem theorem_2_1 (D : Data) (pat : Pattern) (hpat : pat.Holds D)
    {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
    (ξ : ℕ → Ω → E D.d) (hA1 : CompleteRecourse D) (hA2 : A2 P D pat ξ)
    (hA3 : Admissible P D pat ξ) (hX1 : Bornology.IsBounded (X1 D)) :
    ∃ L : ℝ, 0 < L ∧ ∃ α : ℝ, 0 < α ∧ ∃ δ : ℝ, 0 < δ ∧
      ∀ ξ' : ℕ → Ω → E D.d, Admissible P D pat ξ' →
        inputDist P D pat ξ ξ' ≤ ENNReal.ofReal δ →
        value P D pat ξ' ≠ ⊥ → value P D pat ξ' ≠ ⊤ →
        ENNReal.ofReal |(value P D pat ξ).toReal - (value P D pat ξ').toReal| ≤
          ENNReal.ofReal L * (inputDist P D pat ξ ξ' +
            ∑ t ∈ Finset.Icc 2 (D.T - 1), filtDist P D pat ξ ξ' α t) := by sorry

end MultistageStab.Value
Source
Heitsch, Römisch & Strugarek, Stability of multistage stochastic programs, author manuscript (edoc.hu-berlin.de, c. 2005), pp. 4–5, Theorem 2.1, (8)–(9)
Human review
  • Endorsed by Shuze Chen · Oct 7, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 7, 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