Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma 8 (corrected) — E[(b_l − b_l(t))⁺] ≤ K_l √(Σ_{i<t} 1/(T−i−1)²) with K_l = √(Σ_j a_lj² λ_j)

Proved
ResolvingNRM.FRUpper.lemma8_bound_on_b

by mikedeng1 · Oct 6, 2026 · Mathlib 0df444a (Lean v4.33.1)

martingalenetwork-revenue-managementp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1poisson

Consider the network revenue management model with Poisson rates λj>0\lambda_j > 0λj​>0, revenues rj≥0r_j \ge 0rj​≥0, nonnegative consumption matrix A=(alj)A = (a_{lj})A=(alj​), initial capacity C≥0C \ge 0C≥0 and an integer horizon TTT. Run the FR policy with any optimal-solution selector, and let bl=Cl/Tb_l = C_l/Tbl​=Cl​/T and bl(t)=Cl(t)/(T−t)b_l(t) = C_l(t)/(T - t)bl​(t)=Cl​(t)/(T−t), where C(t)C(t)C(t) is the remaining capacity at the start of period ttt. Then for every resource lll and every period ttt with 0≤t≤T−10 \le t \le T - 10≤t≤T−1,

E[(bl−bl(t))+] ≤ Kl∑i=0t−11(T−i−1)2,Kl=∑j=1nalj2λj.\mathbb{E}\big[(b_l - b_l(t))^+\big] \ \le\ K_l \sqrt{\sum_{i=0}^{t-1} \frac{1}{(T - i - 1)^2}}, \qquad K_l = \sqrt{\sum_{j=1}^n a_{lj}^2 \lambda_j}.E[(bl​−bl​(t))+] ≤ Kl​i=0∑t−1​(T−i−1)21​​,Kl​=j=1∑n​alj2​λj​​.

The lemma says the average remaining capacity of FR drifts below its initial value C/TC/TC/T by at most O(1/T−t)O(1/\sqrt{T - t})O(1/T−t​) in expectation; summed over ttt through eq. (33), it gives the O(T)O(\sqrt{T})O(T​) term of Proposition 3.

Formalization Note. The printed lemma has Kl=∑jalj2λj2K_l = \sqrt{\sum_j a_{lj}^2 \lambda_j^2}Kl​=∑j​alj2​λj2​​, which is false: in the proof, eq. (69)–(70) uses E[xj(t)2]≤λj2\mathbb{E}[x_j(t)^2] \le \lambda_j^2E[xj​(t)2]≤λj2​ where the conditional variance of the Poisson(xj(i)x_j(i)xj​(i)) increment is xj(i)≤λjx_j(i) \le \lambda_jxj​(i)≤λj​. Counterexample to the printed form: one class, one resource, a=1a = 1a=1, λ=0.01\lambda = 0.01λ=0.01, C=λTC = \lambda TC=λT, T=1000T = 1000T=1000, t=2t = 2t=2; an exact computation gives E[(b−b(2))+]≈1.96⋅10−5\mathbb{E}[(b - b(2))^+] \approx 1.96 \cdot 10^{-5}E[(b−b(2))+]≈1.96⋅10−5, above the printed bound λ1/9992+1/9982≈1.42⋅10−5\lambda \sqrt{1/999^2 + 1/998^2} \approx 1.42 \cdot 10^{-5}λ1/9992+1/9982​≈1.42⋅10−5 (the corrected bound is 1.42⋅10−41.42 \cdot 10^{-4}1.42⋅10−4). The statement uses the corrected KlK_lKl​. The range t≤T−1t \le T - 1t≤T−1 is where bl(t)b_l(t)bl​(t) is defined; it keeps every T−i−1≥1T - i - 1 \ge 1T−i−1≥1. λj>0\lambda_j > 0λj​>0, r≥0r \ge 0r≥0, alj≥0a_{lj} \ge 0alj​≥0, C≥0C \ge 0C≥0 are the model's standing assumptions (p. 7).

Preamble
import Mathlib
import Definitions.Def_RLPBidPrice_Unbiased_Model
import Definitions.Def_ResolvingNRM_FRUpper_Model

open RLPBidPrice.Unbiased Matrix
Formal statement
namespace ResolvingNRM.FRUpper

/-- Lemma 8 (Bound on `b_l`), p. 43, with the corrected constant `K_l = √(∑_j a_lj² λ_j)`:
`E[(b_l − b_l(t))⁺] ≤ K_l √(∑_{i=0}^{t-1} 1 / (T − i − 1)²)` for every period `t ≤ T − 1`, where
`b_l = C_l / T` and `b_l(t) = C_l(t) / (T − t)` along FR. -/
theorem lemma8_bound_on_b {m n : ℕ} (A : Matrix (Fin m) (Fin n) ℝ) (r lam : Fin n → ℝ)
    (hlam : ∀ j, 0 < lam j) (hr : 0 ≤ r) (hA : ∀ l j, 0 ≤ A l j)
    (sel : (Fin m → ℝ) → (Fin n → ℝ)) (hsel : ResolvingNRM.IRT.IsDLPSelector A r lam sel)
    (T t : ℕ) (ht : t + 1 ≤ T) (C : Fin m → ℝ) (hC : 0 ≤ C) (l : Fin m) :
    frStateExp A r lam sel T t (fun c => max (C l / T - c l / ((T : ℝ) - t)) 0) C
      ≤ Kres A lam l *
          Real.sqrt (∑ i ∈ Finset.range t, 1 / ((T : ℝ) - i - 1) ^ 2) := by sorry

end ResolvingNRM.FRUpper
Source
Bumpensanti, Wang, A Re-solving Heuristic with Uniformly Bounded Loss for Network Revenue Management, arXiv:1802.06192v3, Lemma 8, p. 43 (proof pp. 43–44)
Human review
  • Endorsed by Shuze Chen · Oct 6, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 6, 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