Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

§2.2, display (12) — the omitted-mass recursion

Proved
GaussMixGated.ExactReal.eq_12

by mikedeng1 · Nov 11, 2026 · Mathlib 0df444a (Lean v4.33.1)

distribution-learninggaussian-mixturesmachine-learningp2o-batch-stoc27ap2o-gran-per-chapterp2o-plan-paperp2o-strengthenedp2o-v1probabilitystatistics

Let d≥1d\ge1d≥1, 0<ε,δ<1/40<\varepsilon,\delta<1/40<ε,δ<1/4, Ccov≥1C_{\mathrm{cov}}\ge1Ccov​≥1, and let A∗A_*A∗​, JJJ and β\betaβ be as in (7)–(8). Let b0,b1,…b_0,b_1,\dotsb0​,b1​,… be nonnegative reals with b0≤βb_0\le\betab0​≤β such that, for every j<Jj<Jj<J with bj≤1/2b_j\le1/2bj​≤1/2,

bj+1≤bj+(1−bj){3β+Ccov(2bj+β)}.b_{j+1}\le b_j+(1-b_j)\{3\beta+C_{\mathrm{cov}}(2b_j+\beta)\}.bj+1​≤bj​+(1−bj​){3β+Ccov​(2bj​+β)}.

Then for every 0≤j≤J0\le j\le J0≤j≤J,

bj≤3A∗jβ≤ε2/1000,(12)b_j\le3A_*^j\beta\le\varepsilon^2/1000,\qquad(12)bj​≤3A∗j​β≤ε2/1000,(12)

and in particular bj≤1/2b_j\le1/2bj​≤1/2. In the paper bjb_jbj​ is the latent mass outside the stage-jjj cover, and the recursion is (6) with e=2bj+βe=2b_j+\betae=2bj​+β from (4) with a=βa=\betaa=β. This display keeps the omitted mass small across all J+1J+1J+1 stages.

Differences from the source:

  • generalization: b_j the omitted masses of the algorithm on the success events, C_cov the constant of Theorem 2.3 → any nonnegative sequence with b₀ ≤ β obeying the recursion at stages with b_j ≤ 1/2, any C_cov ≥ 1; reason: the display is an arithmetic consequence of these facts; it specializes to the source's b_j
  • reading: 'This verifies the hypothesis b_j ≤ 1/2 at every stage' → b_j ≤ 1/2 stated as a third conclusion; reason: it is the purpose of the display
Preamble
import Mathlib
import Definitions.Def_GaussMixGated_ExactReal_Schedule
Formal statement
namespace GaussMixGated.ExactReal

/-- §2.2, display (12) (Zheng–Xie–He, STOC 2027 #36, PDF pp. 7–8): the omitted-mass recursion
`b_{j+1} ≤ b_j + (1 − b_j){3β + C_cov(2b_j + β)}` (valid at stages with `b_j ≤ 1/2`), started from
`b₀ ≤ β`, gives `b_j ≤ 3 A*^j β ≤ ε²/1000` for `0 ≤ j ≤ J`; in particular `b_j ≤ 1/2`. -/
theorem eq_12 (d : ℕ) (hd : 1 ≤ d) (ε δ Ccov : ℝ) (hε : 0 < ε) (hε' : ε < 1 / 4)
    (hδ : 0 < δ) (hδ' : δ < 1 / 4) (hC : 1 ≤ Ccov) (b : ℕ → ℝ) (hb : ∀ j, 0 ≤ b j)
    (hb0 : b 0 ≤ betaSched d ε δ Ccov)
    (hrec : ∀ j < Jsched d δ, b j ≤ 1 / 2 →
      b (j + 1) ≤ b j + (1 - b j) *
        (3 * betaSched d ε δ Ccov + Ccov * (2 * b j + betaSched d ε δ Ccov))) :
    ∀ j ≤ Jsched d δ,
      b j ≤ 3 * Astar Ccov ^ j * betaSched d ε δ Ccov ∧
      3 * Astar Ccov ^ j * betaSched d ε δ Ccov ≤ ε ^ 2 / 1000 ∧
      b j ≤ 1 / 2 := by sorry

end GaussMixGated.ExactReal
Source
Zheng, Xie, He, Learning Finite Gaussian Location Mixtures via Gated Moments and Posterior Covers, STOC 2027 submission #36 (October 2026), https://github.com/sigact-stoc/stoc2027/releases/download/papers-1/36.pdf, PDF p. 8 (printed p. 6), §2.2, display (12); recursion on PDF p. 7
Human review
  • Endorsed by Shuze Chen · Nov 12, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Nov 12, 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