Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 7.2.1 — Intrinsic matrix Chernoff

Proved
TroppMatrixConcentration.ch7_intrinsic_chernoff

by tc · Oct 7, 2026 · Mathlib 0df444a (Lean v4.33.1)

matrix-concentrationrandom-matrices

Let XkX_kXk​ be a finite independent family of measurable positive semidefinite complex d×dd\times dd×d random matrices, d≥1d\ge1d≥1, satisfying λmax⁡(Xk)≤L\lambda_{\max}(X_k)\le Lλmax​(Xk​)≤L almost surely, with L>0L>0L>0. Set Y=∑kXkY=\sum_kX_kY=∑k​Xk​. Let M≠0M\ne0M=0 be a semidefinite upper bound for EY\mathbb EYEY, and put r=r(M)r=r(M)r=r(M) and a=λmax⁡(M)a=\lambda_{\max}(M)a=λmax​(M). Then for every θ>0\theta>0θ>0,

Eλmax⁡(Y)≤eθ−1θa+Lθlog⁡(2r).\mathbb E\lambda_{\max}(Y)\le\frac{e^\theta-1}{\theta}a+\frac L\theta\log(2r).Eλmax​(Y)≤θeθ−1​a+θL​log(2r).

For every ε≥L/a\varepsilon\ge L/aε≥L/a,

P{λmax⁡(Y)≥(1+ε)a}≤2r[eε(1+ε)1+ε]a/L.\mathbb P\{\lambda_{\max}(Y)\ge(1+\varepsilon)a\}\le2r\left[\frac{e^\varepsilon}{(1+\varepsilon)^{1+\varepsilon}}\right]^{a/L}.P{λmax​(Y)≥(1+ε)a}≤2r[(1+ε)1+εeε​]a/L.

Lean writes the bracketed power as exp⁡((a/L)(ε−(1+ε)log⁡(1+ε)))\exp((a/L)(\varepsilon-(1+\varepsilon)\log(1+\varepsilon)))exp((a/L)(ε−(1+ε)log(1+ε))), an equal expression on this domain. The nonzero proxy ensures a>0a>0a>0. Both expectation and tail conclusions of the source are included; there is no lower-tail claim. The family may be empty.

The displayed expectation identity EY=∑kEXk\mathbb E Y=\sum_k\mathbb E X_kEY=∑k​EXk​ is also an explicit conclusion.

Preamble
import Definitions.Def_TroppMatrixConcentration_ch7_intrinsic

open MeasureTheory ProbabilityTheory
open scoped Matrix.Norms.L2Operator ComplexOrder

Formal statement
namespace TroppMatrixConcentration

theorem ch7_intrinsic_chernoff {Ω : Type*} [MeasurableSpace Ω]
    (μ : Measure Ω) [IsProbabilityMeasure μ] {d N : ℕ} [NeZero d]
    (X : Fin N → Ω → Matrix (Fin d) (Fin d) ℂ)
    (L : ℝ) (hL : 0 < L) (M : Matrix (Fin d) (Fin d) ℂ) (hM : M ≠ 0)
    (hMeas : ∀ k, Measurable (X k)) (hIndep : iIndepFun X μ)
    (hPSD : ∀ k, ∀ᵐ ω ∂μ, (X k ω).PosSemidef)
    (hBound : ∀ k, ∀ᵐ ω ∂μ, lambdaMax (X k ω) ≤ L)
    (hMeanBound : loewnerLE (∫ ω, ∑ k, X k ω ∂μ) M) :
    let Y := fun ω => ∑ k, X k ω
    let r := intrinsicDimension M
    let a := lambdaMax M
    (∫ ω, Y ω ∂μ) = ∑ k, ∫ ω, X k ω ∂μ ∧
    (∀ θ : ℝ, 0 < θ → (∫ ω, lambdaMax (Y ω) ∂μ) ≤
      (Real.exp θ - 1) / θ * a + L / θ * Real.log (2 * r)) ∧
    ∀ ε : ℝ, L / a ≤ ε →
      (μ {ω | (1 + ε) * a ≤ lambdaMax (Y ω)}).toReal ≤
        2 * r * Real.exp (a / L * (ε - (1 + ε) * Real.log (1 + ε))) := by sorry

end TroppMatrixConcentration
Source
Joel A. Tropp, An Introduction to Matrix Concentration Inequalities, arXiv:1501.01571v1 (7 January 2015); https://arxiv.org/abs/1501.01571v1; Theorem 7.2.1, equations (7.2.1–2), printed pp. 106–107; standing independence confirmed in proof, Section 7.6, printed p. 113.
Read-back

What the Lean code literally says, in plain math · gpt-6

For every measurable space Ω\OmegaΩ, every probability measure μ\muμ on it, natural numbers d,Nd,Nd,N with d≠0d\ne0d=0, a family Xk:Ω→Cd×dX_k:\Omega\to\mathbb C^{d\times d}Xk​:Ω→Cd×d indexed by k∈{0,…,N−1}k\in\{0,\ldots,N-1\}k∈{0,…,N−1}, a real number L>0L>0L>0, and a nonzero matrix M∈Cd×dM\in\mathbb C^{d\times d}M∈Cd×d, assume that every XkX_kXk​ is measurable for the Borel measurable structure on matrices, that the family (Xk)k(X_k)_k(Xk​)k​ is mutually independent under μ\muμ, that for every kkk the matrix Xk(ω)X_k(\omega)Xk​(ω) is positive semidefinite for μ\muμ-almost every ω\omegaω, and that for every kkk one has λ(Xk(ω))≤L\lambda(X_k(\omega))\le Lλ(Xk​(ω))≤L for μ\muμ-almost every ω\omegaω. Here λ(A)=sup⁡spec⁡R(A)\lambda(A)=\sup\operatorname{spec}_{\mathbb R}(A)λ(A)=supspecR​(A) is the supremum in R\mathbb RR of the spectrum of the complex matrix AAA regarded as an element of an algebra over R\mathbb RR; equivalently its spectrum consists of real scalars ttt for which tI−AtI-AtI−A is not invertible. Assume also that M−∫Ω∑k=0N−1Xk(ω) dμ(ω)M-\int_\Omega\sum_{k=0}^{N-1}X_k(\omega)\,d\mu(\omega)M−∫Ω​∑k=0N−1​Xk​(ω)dμ(ω) is positive semidefinite. Set Y(ω)=∑k=0N−1Xk(ω)Y(\omega)=\sum_{k=0}^{N-1}X_k(\omega)Y(ω)=∑k=0N−1​Xk​(ω), r=Re⁡(tr⁡M)/∥M∥2→2r=\operatorname{Re}(\operatorname{tr}M)/\lVert M\rVert_{2\to2}r=Re(trM)/∥M∥2→2​, where the denominator is the operator norm for Euclidean norms, and a=λ(M)a=\lambda(M)a=λ(M). The assertion is the conjunction of the following three conclusions:

∫ΩY(ω) dμ(ω)=∑k=0N−1∫ΩXk(ω) dμ(ω);\int_\Omega Y(\omega)\,d\mu(\omega)=\sum_{k=0}^{N-1}\int_\Omega X_k(\omega)\,d\mu(\omega);∫Ω​Y(ω)dμ(ω)=k=0∑N−1​∫Ω​Xk​(ω)dμ(ω);

for every real θ>0\theta>0θ>0,

∫Ωλ(Y(ω)) dμ(ω)≤exp⁡(θ)−1θ a+Lθlog⁡(2r);\int_\Omega\lambda(Y(\omega))\,d\mu(\omega)\le\frac{\exp(\theta)-1}{\theta}\,a+\frac L\theta\log(2r);∫Ω​λ(Y(ω))dμ(ω)≤θexp(θ)−1​a+θL​log(2r);

and for every real ε\varepsilonε satisfying L/a≤εL/a\le\varepsilonL/a≤ε,

μ{ω∈Ω:(1+ε)a≤λ(Y(ω))}≤2rexp⁡ ⁣(aL[ε−(1+ε)log⁡(1+ε)]).\mu\{\omega\in\Omega:(1+\varepsilon)a\le\lambda(Y(\omega))\}\le 2r\exp\!\left(\frac aL\bigl[\varepsilon-(1+\varepsilon)\log(1+\varepsilon)\bigr]\right).μ{ω∈Ω:(1+ε)a≤λ(Y(ω))}≤2rexp(La​[ε−(1+ε)log(1+ε)]).

The measure on the left of the last inequality is converted from its extended nonnegative real value to a real number; it is finite since μ\muμ is a probability measure. All integrals are Bochner integrals. The quantification includes N=0N=0N=0, when the sums and YYY are zero and the assumptions on the family and its independence are vacuous, while the mean bound requires MMM to be positive semidefinite. There is no separate hypothesis declaring MMM Hermitian or positive semidefinite, no separate integrability hypothesis, and no separate positivity hypothesis for aaa or rrr; these quantities are used exactly as defined above. The definitions are total: real division by zero is zero, the real logarithm at zero is zero and at a negative real number is the logarithm of its absolute value, the real supremum of the empty set is zero, and a nonintegrable function has Bochner integral zero. The displayed statement uses the non-strict tail event and the non-strict threshold condition exactly as written.

Human review
  • Endorsed by Shuze Chen · Oct 7, 2026

    Confirmed by the moderator at approval.

  • Endorsed by tc · Oct 7, 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