Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 2, p. 317 — the BCD computes a (1 + ε)n-rounding ellipsoid of m points in ℝⁿ within O(n(1/ε + ln n + ln ln m)) iterations

Proved
KhachiyanRound.BCD.theorem_2

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

ellipsoiditeration-complexitylowner-johnp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1rounding

There is a universal constant C>0C > 0C>0 with the following property. Let n≥1n \ge 1n≥1, let A={a1,…,am}⊂Rn\mathcal A = \{a_1,\dots,a_m\} \subset \mathbb{R}^nA={a1​,…,am​}⊂Rn have affine hull Rn\mathbb{R}^nRn, and let ε>0\varepsilon > 0ε>0. Run the BCD method on the centrally symmetric lifted set A′={±(aj;1)}⊂Rn+1\mathcal A' = \{\pm(a_j;1)\} \subset \mathbb{R}^{n+1}A′={±(aj​;1)}⊂Rn+1 of (3.3) — with any tie-breaking rule — producing iterates p0,p1,…p_0, p_1, \dotsp0​,p1​,…, and let pj=pj++pj−p_j = p_j^+ + p_j^-pj​=pj+​+pj−​ denote aggregated weights. Then there is an iterate pkp_kpk​ with

k≤C (n+1)(1ε+ln⁡(n+1)+ln⁡ln⁡(2m))k \le C\,(n+1)\Big(\frac1\varepsilon + \ln(n+1) + \ln\ln(2m)\Big)k≤C(n+1)(ε1​+ln(n+1)+lnln(2m))

that satisfies the relaxed optimality conditions (2.10) on A′\mathcal A'A′ with ε′=nn+1ε\varepsilon' = \frac{n}{n+1}\varepsilonε′=n+1n​ε (3.2), whose moment matrix is positive definite, and for which the ellipsoid

E={x∈Rn  ∣  (x;1)∈1+(1+ε)n  Epk′}E = \big\{x \in \mathbb{R}^n \;\big|\; (x;1) \in \sqrt{1+(1+\varepsilon)n}\;E'_{p_k}\big\}E={x∈Rn​(x;1)∈1+(1+ε)n​Epk​′​}

of Lemma 5 is a (1+ε)n(1+\varepsilon)n(1+ε)n-rounding of A\mathcal AA (1.2):

[(1+ε)n]−1E⊆conv⁡(A)⊆E,[(1+\varepsilon)n]^{-1}E \subseteq \operatorname{conv}(\mathcal A) \subseteq E,[(1+ε)n]−1E⊆conv(A)⊆E,

the shrinking being about the centre b=∑jpjajb = \sum_j p_j a_jb=∑j​pj​aj​ of EEE.

This is the paper's main result: a simple first-order method computes a near-optimal rounding ellipsoid of an arbitrary polytope given by its vertices, in a number of iterations that depends on mmm only through ln⁡ln⁡m\ln\ln mlnlnm.

Formalization Note The page states N(ε)=O(mn2(ε−1+ln⁡n+ln⁡ln⁡m))N(\varepsilon) = O(mn^2(\varepsilon^{-1} + \ln n + \ln\ln m))N(ε)=O(mn2(ε−1+lnn+lnlnm)) arithmetic operations and comparisons. The Lean statement bounds the index of an iterate that passes the stopping test; the cost per iteration (O(nm)O(nm)O(nm), from the rank-one correction) and the O(mn2)O(mn^2)O(mn2) cost of forming A0−1A_0^{-1}A0−1​ are not formalized. The bound is written in the lifted parameters (n+1,2m)(n+1, 2m)(n+1,2m), which keeps it positive for every n≥1n\ge1n≥1 (at n=1n = 1n=1, m=2m = 2m=2 the page's ln⁡n+ln⁡ln⁡m\ln n + \ln\ln mlnn+lnlnm is negative); wherever the page's expression is meaningful the two agree up to the constant. The constant CCC is quantified before every instance datum. Ep′E'_{p}Ep′​ is the ellipsoid (3.4) of the aggregated weights.


Retired 2026-10-27: a Lean-vs-paper fidelity audit found this statement weaker than, or narrower than, the cited result. A corrected statement will be resubmitted.

Preamble
import Mathlib
import Definitions.Def_LinearOptimization_Ellipsoid
import Definitions.Def_KhachiyanRound_BCD_Setting

open scoped Pointwise
Formal statement
namespace KhachiyanRound.BCD

theorem theorem_2 :
    ∃ C : ℝ, 0 < C ∧ ∀ (n m : ℕ) (a : Fin m → Fin n → ℝ), 1 ≤ n →
      affineSpan ℝ (Set.range a) = ⊤ → ∀ ε : ℝ, 0 < ε →
        ∀ p : ℕ → (Fin m ⊕ Fin m → ℝ), IsBCDRun (liftPts a) p →
          ∃ k : ℕ, (k : ℝ) ≤ C * (n + 1) *
              (1 / ε + Real.log (n + 1) + Real.log (Real.log (2 * m))) ∧
            IsRelaxedOpt (liftPts a) (p k) ((n : ℝ) / (n + 1) * ε) ∧
            (momentMatrix (liftPts a) (p k)).PosDef ∧
            (AffineMap.homothety (∑ j, aggr (p k) j • a j) ((1 + ε) * (n : ℝ))⁻¹) ''
                lemma5Ellipsoid a (aggr (p k)) ε ⊆ convexHull ℝ (Set.range a) ∧
            convexHull ℝ (Set.range a) ⊆ lemma5Ellipsoid a (aggr (p k)) ε := by sorry
end KhachiyanRound.BCD
Source
Khachiyan, Rounding of polytopes in the real number model of computation, Math. Oper. Res. 21 (1996), p. 317, Theorem 2; (1.2), p. 307; (3.2)–(3.4), p. 315
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