Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma C.4 — exponential moment of the one-window OPAC deviation

Proved
ChoicePAC.MidPoint.lemma_C4_window_mgf

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

moment-generating-functionp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1poisson

Consider a valid instance. There is a constant φ\varphiφ such that the following holds for every k≥1k \ge 1k≥1, every window [s,t)[s,t)[s,t) of length ℓ=t−s≥0\ell = t-s \ge 0ℓ=t−s≥0, every admissible allocation ppp (with pj≥0p_j \ge 0pj​≥0 and ∑j∈Sqpj≤1\sum_{j\in S_q}p_j \le 1∑j∈Sq​​pj​≤1), every rrr with ∣rξmax⁡∣≤1|r\xi_{\max}| \le 1∣rξmax​∣≤1 and every resource iii. When OPAC is applied during [s,t)[s,t)[s,t) with allocation probabilities ppp,

E[exp⁡(rΔ~i(s,t))]≤∏qexp⁡(kλq(t−s)r2∑j∈Sqpj E[Aij2])≤exp⁡(kφ(t−s)r2),\mathbb E\Big[\exp\big(r\tilde\Delta_i(s,t)\big)\Big] \le \prod_q \exp\Big(k\lambda_q(t-s)r^2\sum_{j\in S_q}p_j\,\mathbb E[A_{ij}^2]\Big) \le \exp\big(k\varphi(t-s)r^2\big),E[exp(rΔ~i​(s,t))]≤q∏​exp(kλq​(t−s)r2j∈Sq​∑​pj​E[Aij2​])≤exp(kφ(t−s)r2),

where Δ~i(s,t)=ζi−∑jkλq(j)(t−s)pjAˉij\tilde\Delta_i(s,t) = \zeta_i - \sum_j k\lambda_{q(j)}(t-s)p_j\bar A_{ij}Δ~i​(s,t)=ζi​−∑j​kλq(j)​(t−s)pj​Aˉij​ is the deviation of the consumption of resource iii from its conditional mean.

This sub-Gaussian bound on the window deviation is the basic probabilistic estimate behind the bound on the hitting time in the proof of Theorem 5.3.

Formalization Note φ\varphiφ is chosen before kkk, ℓ\ellℓ, ppp, rrr and iii: the page says "independent of iii", and the proof's φ\varphiφ does not depend on the others either. The page's first equality (the definition of Δ~\tilde\DeltaΔ~ via i.i.d. copies) is the definition of the expectation used here; conditioning on {pj}\{p_j\}{pj​} is the choice of a fixed ppp. The standing hypotheses (IsValid) are the disclosed reading of Sec. 2.1, p. 315, Sec. 2.3, p. 317 and Sec. 5.2, p. 327: printed are λq≥0\lambda_q \ge 0λq​≥0, bounded nonnegative consumption and rj(0)=0r_j(0)=0rj​(0)=0; read in are C≥0C \ge 0C≥0, consumption laws that are probability measures, Aij≤ξjA_{ij} \le \xi_jAij​≤ξj​ almost surely (the page writes ξj≥Aij\xi_j \ge A_{ij}ξj​≥Aij​ on p. 317 and Aij<ξjA_{ij} < \xi_jAij​<ξj​ on p. 327; the weaker ≤\le≤ is used), measurable revenue functions, and revenue that is nonnegative and bounded by a common constant (implicit on the page, needed in App. C.4).

Preamble
import Mathlib
import Definitions.Def_ChoicePAC_MidPoint_Model
import Definitions.Def_ChoicePAC_MidPoint_OPAC
Formal statement
namespace ChoicePAC.MidPoint
open MeasureTheory in
theorem lemma_C4_window_mgf {NT n m : ℕ} (I : Instance NT n m) (hI : I.IsValid) :
    ∃ φ : ℝ, ∀ k : ℕ, 1 ≤ k → ∀ ℓ : ℝ, 0 ≤ ℓ → ∀ p : Fin n → ℝ, I.AdmissibleProbs p →
      ∀ r : ℝ, |r * I.ξmax| ≤ 1 → ∀ i : Fin m,
        I.opacExpect k ℓ p (fun ζ => Real.exp (r * (ζ i - I.opacMean k ℓ p i))) ≤
            ∏ q, Real.exp ((k : ℝ) * I.lam q * ℓ * r ^ 2 *
              ∑ j ∈ I.S q, p j * ∫ a, (a i) ^ 2 ∂(I.D j)) ∧
          ∏ q, Real.exp ((k : ℝ) * I.lam q * ℓ * r ^ 2 *
              ∑ j ∈ I.S q, p j * ∫ a, (a i) ^ 2 ∂(I.D j)) ≤
            Real.exp ((k : ℝ) * φ * ℓ * r ^ 2) := by sorry
end ChoicePAC.MidPoint
Source
Jasin, Kumar, A Re-Solving Heuristic with Bounded Revenue Loss for Network Revenue Management with Customer Choice, Math. Oper. Res. 37(2), 2012, App. C.2, p. 338, Lemma C.4
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