Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Corollary 6.1.2 — matrix Bernstein for uncentered summands

Proved
IntroMatrixConc.Sampling.corollary_6_1_2

by mikedeng1 · Oct 9, 2026 · Mathlib 0df444a (Lean v4.33.1)

matrix-bernsteinmatrix-concentrationp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-bookp2o-v1random-matrices

Let S1,…,SNS_1,\ldots,S_NS1​,…,SN​ be independent measurable complex d1×d2d_1\times d_2d1​×d2​ random matrices on a probability space, with d1,d2≥1d_1,d_2\ge1d1​,d2​≥1. Suppose each has a finite expectation and its deviation satisfies ∥Sk−ESk∥≤L\|S_k-\mathbb E S_k\|\le L∥Sk​−ESk​∥≤L almost surely, where L≥0L\ge0L≥0. Set Z=∑kSkZ=\sum_kS_kZ=∑k​Sk​ and

v(Z)=max⁡{∥E[(Z−EZ)(Z−EZ)∗]∥,∥E[(Z−EZ)∗(Z−EZ)]∥}.v(Z)=\max\{\|\mathbb E[(Z-\mathbb EZ)(Z-\mathbb EZ)^*]\|,\|\mathbb E[(Z-\mathbb EZ)^*(Z-\mathbb EZ)]\|\}.v(Z)=max{∥E[(Z−EZ)(Z−EZ)∗]∥,∥E[(Z−EZ)∗(Z−EZ)]∥}.

Then v(Z)v(Z)v(Z) also equals the maximum of the norms of the corresponding sums of centered second moments. Moreover,

E∥Z−EZ∥≤2v(Z)log⁡(d1+d2)+Llog⁡(d1+d2)3,\mathbb E\|Z-\mathbb EZ\|\le\sqrt{2v(Z)\log(d_1+d_2)}+\frac{L\log(d_1+d_2)}{3},E∥Z−EZ∥≤2v(Z)log(d1​+d2​)​+3Llog(d1​+d2​)​,

and, for every t≥0t\ge0t≥0,

P{∥Z−EZ∥≥t}≤(d1+d2)exp⁡ ⁣(−t2/2v(Z)+Lt/3).\mathbb P\{\|Z-\mathbb EZ\|\ge t\}\le(d_1+d_2)\exp\!\left(-\frac{t^2/2}{v(Z)+Lt/3}\right).P{∥Z−EZ∥≥t}≤(d1​+d2​)exp(−v(Z)+Lt/3t2/2​).

This version applies matrix Bernstein directly to summands that need not have zero mean.

Formalization Note Finite expectations are explicit integrability hypotheses, following the book's regularity convention in §2.2.1. At a zero denominator, the displayed exponential uses Lean's total division convention; the resulting bound is harmless because the centered sum is then zero almost surely.

Preamble
import Definitions.Def_IntroMatrixConc_Sampling_Defs

open MeasureTheory ProbabilityTheory TroppMatrixConcentration
open scoped Matrix.Norms.L2Operator

noncomputable section
Formal statement
namespace IntroMatrixConc.Sampling

/-- Tropp, Corollary 6.1.2, p. 77 (PDF p. 83). -/
theorem corollary_6_1_2 {Ω : Type*} [MeasurableSpace Ω]
    (μ : Measure Ω) [IsProbabilityMeasure μ]
    {d₁ d₂ N : ℕ} [NeZero d₁] [NeZero d₂]
    (S : Fin N → Ω → Matrix (Fin d₁) (Fin d₂) ℂ)
    (L : ℝ) (hL : 0 ≤ L)
    (hMeas : ∀ k, Measurable (S k))
    (hInt : ∀ k, Integrable (S k) μ)
    (hIndep : iIndepFun S μ)
    (hBound : ∀ k, ∀ᵐ ω ∂μ,
      spectralNorm (S k ω - ∫ ω, S k ω ∂μ) ≤ L) :
    let Z := fun ω => ∑ k, S k ω
    let V := rectSecondMoment μ (fun ω => Z ω - ∫ ω, Z ω ∂μ)
    V = max
      (spectralNorm (∑ k, ∫ ω,
        (S k ω - ∫ η, S k η ∂μ) * (S k ω - ∫ η, S k η ∂μ).conjTranspose ∂μ))
      (spectralNorm (∑ k, ∫ ω,
        (S k ω - ∫ η, S k η ∂μ).conjTranspose * (S k ω - ∫ η, S k η ∂μ) ∂μ)) ∧
    (∫ ω, spectralNorm (Z ω - ∫ η, Z η ∂μ) ∂μ) ≤
      Real.sqrt (2 * V * Real.log (d₁ + d₂)) + L * Real.log (d₁ + d₂) / 3 ∧
    ∀ t : ℝ, 0 ≤ t →
      (μ {ω | t ≤ spectralNorm (Z ω - ∫ η, Z η ∂μ)}).toReal ≤
        (d₁ + d₂ : ℕ) * Real.exp (-(t ^ 2 / 2) / (V + L * t / 3)) := by sorry

end IntroMatrixConc.Sampling
Source
Tropp, arXiv:1501.01571v1, Corollary 6.1.2, p. 77 (PDF p. 83)
Human review
  • Endorsed by Shuze Chen · Oct 9, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 9, 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