Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 4, p. 30 — (NV-reg) is uniformly stable with parameter (b∨h)²X²_max/(2nλ)

Proved
BigDataNV.Reg.theorem4_uniform_stability

by mikedeng1 · Oct 5, 2026 · Mathlib 0df444a (Lean v4.33.1)

newsvendorp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1regularizationuniform-stability

Let b,h>0b,h>0b,h>0 be the newsvendor costs, λ>0\lambda>0λ>0 the regularization parameter and Xmax⁡≥0X_{\max}\ge0Xmax​≥0. Let Sn={(xj,dj)}j=1nS_n=\{(x_j,d_j)\}_{j=1}^nSn​={(xj​,dj​)}j=1n​ be a sample whose feature vectors satisfy ∥xj∥22≤Xmax⁡2\|x_j\|_2^2\le X_{\max}^2∥xj​∥22​≤Xmax2​, and fix an index iii. Let q^\hat qq^​ minimize the (NV-reg) objective

1n∑j=1nC(q⊤xj;dj)+λ∥q∥22\frac1n\sum_{j=1}^n C(q^\top x_j;d_j)+\lambda\|q\|_2^2n1​j=1∑n​C(q⊤xj​;dj​)+λ∥q∥22​

over q∈Rpq\in\mathbb R^pq∈Rp, and let q^∖i\hat q^{\setminus i}q^​∖i minimize the leave-one-out objective 1n∑j≠iC(q⊤xj;dj)+λ∥q∥22\frac1n\sum_{j\ne i} C(q^\top x_j;d_j)+\lambda\|q\|_2^2n1​∑j=i​C(q⊤xj​;dj​)+λ∥q∥22​. Then for every feature vector xxx with ∥x∥22≤Xmax⁡2\|x\|_2^2\le X_{\max}^2∥x∥22​≤Xmax2​ and every demand ddd,

∣C(q^⊤x;d)−C((q^∖i)⊤x;d)∣≤(b∨h)2Xmax⁡22nλ.\bigl|C(\hat q^\top x;d)-C((\hat q^{\setminus i})^\top x;d)\bigr|\le\frac{(b\vee h)^2X_{\max}^2}{2n\lambda}.​C(q^​⊤x;d)−C((q^​∖i)⊤x;d)​≤2nλ(b∨h)2Xmax2​​.

This is the uniform stability of (NV-reg) with parameter αnr=(b∨h)2Xmax⁡2/(2nλ)\alpha_n^r=(b\vee h)^2X_{\max}^2/(2n\lambda)αnr​=(b∨h)2Xmax2​/(2nλ), display (28). Together with the generalization bound for stable algorithms (Theorem 6), it yields the feature-count-free bound of Theorem 2.

Formalization Note The page writes the constant as (b∨h)2/(2Xmax⁡−2)⋅1/(nλ)(b\vee h)^2/(2X_{\max}^{-2})\cdot 1/(n\lambda)(b∨h)2/(2Xmax−2​)⋅1/(nλ), i.e. a factor Xmax⁡2X_{\max}^2Xmax2​. The leave-one-out problem keeps the weight 1/n1/n1/n, as in Theorem 5 (Bousquet and Elisseeff's (20)); running (NV-reg) literally on n−1n-1n−1 points would use 1/(n−1)1/(n-1)1/(n−1). The statement holds for any pair of minimizers. Symmetry of the algorithm is not stated, as the objective does not depend on the order of the sample. No bound on the demands is needed.

Preamble
import Mathlib
import Definitions.Def_BigDataNV_Reg_Setting
Formal statement
namespace BigDataNV.Reg

/-- Theorem 4, p. 30: (NV-reg) is uniformly stable with respect to the newsvendor cost, with
stability parameter `(b ∨ h)² X²_max / (2 n λ)`, in the form used by Theorem 5 (p. 31): the
comparison solution minimizes the leave-one-out objective with weight `1/n`. -/
theorem theorem4_uniform_stability {p n : ℕ} (b h lam Xmax : ℝ)
    (hb : 0 < b) (hh : 0 < h) (hlam : 0 < lam) (hX : 0 ≤ Xmax)
    (S : Fin n → EuclideanSpace ℝ (Fin p) × ℝ) (hS : ∀ j, ‖(S j).1‖ ^ 2 ≤ Xmax ^ 2)
    (i : Fin n) (q q' : EuclideanSpace ℝ (Fin p))
    (hq : IsNVRegSolution b h lam S q) (hq' : IsNVRegLooSolution b h lam S i q') :
    ∀ x : EuclideanSpace ℝ (Fin p), ‖x‖ ^ 2 ≤ Xmax ^ 2 → ∀ d : ℝ,
      |nvCost b h (inner ℝ q x) d - nvCost b h (inner ℝ q' x) d| ≤
        (max b h) ^ 2 * Xmax ^ 2 / (2 * (n : ℝ) * lam) := by sorry

end BigDataNV.Reg
Source
Rudin & Vahn, The Big Data Newsvendor: Practical Insights from Machine Learning, MIT Sloan Working Paper 5036-13 (version of February 6, 2014), p. 30, Theorem 4, display (28); Definition 1, p. 28, display (24); proof p. 31
Human review
  • Endorsed by Shuze Chen · Oct 5, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 5, 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