Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 3.5 — IMPROVE-COVER stops within 2+20ρε−3ln⁡(4m/ε)2+20\rho\varepsilon^{-3}\ln(4m/\varepsilon)2+20ρε−3ln(4m/ε) calls, and 2+196ρε−2ln⁡(4m/ε)2+196\rho\varepsilon^{-2}\ln(4m/\varepsilon)2+196ρε−2ln(4m/ε) from a 6ε6\varepsilon6ε-optimal start

Proved
FracPackCover.Covering.theorem_3_5

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

fractional-coveringp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1potential-functionrunning-time

Let A≥0A\ge0A≥0, b>0b>0b>0, PPP be covering data, ρ>0\rho>0ρ>0 a width bound, and let subroutine (7) be an exact maximization oracle. Let 0<ε<10<\varepsilon<10<ε<1 and x∈Px\in Px∈P with λ(x)>0\lambda(x)>0λ(x)>0.

  1. IMPROVE-COVER(x,ε)(x,\varepsilon)(x,ε) terminates, and it makes at most
2+20 ρ ε−3ln⁡4mε2+20\,\rho\,\varepsilon^{-3}\ln\frac{4m}{\varepsilon}2+20ρε−3lnε4m​

calls to subroutine (7); the number of evaluations of its while-test obeys the same bound. 2. If moreover ε≤112\varepsilon\le\frac1{12}ε≤121​ and the initial solution is 6ε6\varepsilon6ε-optimal, i.e. (1−6ε)λ(x′)≤λ(x)(1-6\varepsilon)\lambda(x')\le\lambda(x)(1−6ε)λ(x′)≤λ(x) for every x′∈Px'\in Px′∈P, then IMPROVE-COVER(x,ε)(x,\varepsilon)(x,ε) terminates after at most

2+196 ρ ε−2ln⁡4mε2+196\,\rho\,\varepsilon^{-2}\ln\frac{4m}{\varepsilon}2+196ρε−2lnε4m​

calls (and as many while-tests).

These are the per-call costs from which the total cost of the covering algorithm (Theorem 3.7) is assembled: the first bound for the calls with ε=1/6\varepsilon=1/6ε=1/6, the second for the ε\varepsilonε-scaling phases.

Formalization Note The paper writes O(ε−3ρlog⁡(mε−1))O(\varepsilon^{-3}\rho\log(m\varepsilon^{-1}))O(ε−3ρlog(mε−1)) and O(ε−2ρlog⁡(mε−1))O(\varepsilon^{-2}\rho\log(m\varepsilon^{-1}))O(ε−2ρlog(mε−1)) iterations; its proof yields at most 1+20ρε−3ln⁡(4m/ε)1+20\rho\varepsilon^{-3}\ln(4m/\varepsilon)1+20ρε−3ln(4m/ε) and 1+196ρε−2ln⁡(4m/ε)1+196\rho\varepsilon^{-2}\ln(4m/\varepsilon)1+196ρε−2ln(4m/ε) updates, and one more oracle call for the final test. The second sentence of the printed theorem names IMPROVE-PACKING, a misprint for IMPROVE-COVER, which is what is stated here. Termination is asserted: the run with that many while-tests returns a result.

Preamble
import Mathlib
import Definitions.Def_FracPackCover_Covering_Basic
import Definitions.Def_FracPackCover_Covering_ImproveCover
Formal statement
namespace FracPackCover.Covering

/-- Theorem 3.5 (Plotkin–Shmoys–Tardos, Cornell ORIE TR 999, p. 20), with the explicit constants
of its proof. Let `0 < ε < 1`, `x ∈ P` with `λ(x) > 0`, `ρ` a width bound and `orc` an exact
maximization oracle.
1. IMPROVE-COVER(x, ε) stops within `B₁ = 2 + 20 ρ ε⁻³ ln(4 m ε⁻¹)` evaluations of its while-test
   and makes at most `B₁` oracle calls (the paper writes `O(ε⁻³ ρ log(m ε⁻¹))` iterations).
2. If moreover `ε ≤ 1/12` and `x` is `6ε`-optimal (`(1 − 6ε) λ(x') ≤ λ(x)` for every `x' ∈ P`), it
   stops within `B₂ = 2 + 196 ρ ε⁻² ln(4 m ε⁻¹)` tests and makes at most `B₂` oracle calls (the paper
   writes `O(ε⁻² ρ log(m ε⁻¹))`; its second sentence names IMPROVE-PACKING, a misprint for
   IMPROVE-COVER). -/
theorem theorem_3_5 {m n : ℕ} [NeZero m] (A : Matrix (Fin m) (Fin n) ℝ) (b : Fin m → ℝ)
    (P : Set (Fin n → ℝ)) (hdata : IsCoveringData A b P) (ρ : ℝ) (hρ : WidthBound A b P ρ)
    (orc : (Fin m → ℝ) → (Fin n → ℝ)) (horc : IsMaxOracle A P orc)
    (x : Fin n → ℝ) (hx : x ∈ P) (hlam : 0 < lam A b x) (ε : ℝ) (hε : 0 < ε) (hε1 : ε < 1) :
    (∃ x' : Fin n → ℝ, ∃ c : ℕ,
      improveCover A b orc ρ ε x ⌈2 + 20 * ρ * ε⁻¹ ^ 3 * Real.log (4 * m * ε⁻¹)⌉₊ = some (x', c) ∧
      (c : ℝ) ≤ 2 + 20 * ρ * ε⁻¹ ^ 3 * Real.log (4 * m * ε⁻¹)) ∧
    (ε ≤ 1 / 12 → (∀ x' ∈ P, (1 - 6 * ε) * lam A b x' ≤ lam A b x) →
      ∃ x' : Fin n → ℝ, ∃ c : ℕ,
        improveCover A b orc ρ ε x ⌈2 + 196 * ρ * ε⁻¹ ^ 2 * Real.log (4 * m * ε⁻¹)⌉₊ =
          some (x', c) ∧
        (c : ℝ) ≤ 2 + 196 * ρ * ε⁻¹ ^ 2 * Real.log (4 * m * ε⁻¹)) := by sorry

end FracPackCover.Covering
Source
Plotkin, Shmoys, Tardos, Fast Approximation Algorithms for Fractional Packing and Covering Problems, Cornell ORIE Tech. Rep. 999 (1992), p. 20, Theorem 3.5
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