Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 4.6.1 — Hermitian Gaussian and Rademacher series

Proved
TroppMatrixConcentration.ch4_hermitian_series

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

matrix-concentrationrandom-matrices

Let AkA_kAk​ be a finite family of fixed Hermitian complex d×dd\times dd×d matrices, with d≥1d\ge1d≥1. On an arbitrary probability space, let gkg_kgk​ be independent measurable real random variables, either all standard normal or all Rademacher. Put Y=∑kgkAkY=\sum_k g_k A_kY=∑k​gk​Ak​ and v=∥EY2∥v=\|\mathbb E Y^2\|v=∥EY2∥. Then

v=∥∑kAk2∥,Eλmax⁡(Y)≤2vlog⁡d,v=\left\|\sum_k A_k^2\right\|,\qquad\mathbb E\lambda_{\max}(Y)\le\sqrt{2v\log d},v=​k∑​Ak2​​,Eλmax​(Y)≤2vlogd​, P{λmax⁡(Y)≥t}≤dexp⁡(−t2/(2v))(t≥0).\mathbb P\{\lambda_{\max}(Y)\ge t\}\le d\exp(-t^2/(2v))\qquad(t\ge0).P{λmax​(Y)≥t}≤dexp(−t2/(2v))(t≥0).

At v=0v=0v=0 the tail right side is ddd for t=0t=0t=0 and zero for t>0t>0t>0. Empty finite families are allowed. The norm is the Euclidean operator norm; the expectation concerns the largest eigenvalue rather than the norm.

Preamble
import Definitions.Def_TroppMatrixConcentration_ch4_scalar_laws
import Definitions.Def_TroppMatrixConcentration_dilation

open MeasureTheory ProbabilityTheory
open scoped Matrix.Norms.L2Operator

Formal statement
namespace TroppMatrixConcentration

theorem ch4_hermitian_series {Ω : Type*} [MeasurableSpace Ω]
    (μ : Measure Ω) [IsProbabilityMeasure μ] {d N : ℕ} [NeZero d]
    (A : Fin N → Matrix (Fin d) (Fin d) ℂ) (hA : ∀ k, (A k).IsHermitian)
    (g : Fin N → Ω → ℝ) (hMeas : ∀ k, Measurable (g k))
    (hIndep : iIndepFun g μ)
    (hLaw : (∀ k, standardGaussianLaw μ (g k)) ∨ (∀ k, rademacherLaw μ (g k))) :
    let Y := fun ω => ∑ k, g k ω • A k
    let v := hermitianSecondMoment μ Y
    v = spectralNorm (∑ k, A k ^ 2) ∧
    (∫ ω, lambdaMax (Y ω) ∂μ) ≤ Real.sqrt (2 * v * Real.log d) ∧
    ∀ t : ℝ, 0 ≤ t → (μ {ω | t ≤ lambdaMax (Y ω)}).toReal ≤
      gaussianSeriesTail d v t := 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 4.6.1, equations (4.6.1–3), printed p. 51.
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, every positive integer ddd, every nonnegative integer NNN, every family A0,…,AN−1A_0,\ldots,A_{N-1}A0​,…,AN−1​ of complex d×dd\times dd×d matrices satisfying Ak∗=AkA_k^*=A_kAk∗​=Ak​ for every kkk, and every family of measurable functions gk:Ω→Rg_k:\Omega\to\mathbb Rgk​:Ω→R that are jointly independent under μ\muμ, assume that either every pushforward measure (gk)∗μ(g_k)_*\mu(gk​)∗​μ is the real Gaussian measure with mean 000 and variance 111, or every such pushforward is 12δ1+12δ−1\tfrac12\delta_1+\tfrac12\delta_{-1}21​δ1​+21​δ−1​; the disjunction requires a common choice of law for the whole family. Define Y(ω)=∑k=0N−1gk(ω)AkY(\omega)=\sum_{k=0}^{N-1}g_k(\omega)A_kY(ω)=∑k=0N−1​gk​(ω)Ak​ and v=∥∫ΩY(ω)2 dμ(ω)∥opv=\left\|\int_\Omega Y(\omega)^2\,d\mu(\omega)\right\|_{\mathrm{op}}v=​∫Ω​Y(ω)2dμ(ω)​op​, where squares are matrix products, the matrix integral is the Bochner integral, and ∥⋅∥op\|\cdot\|_{\mathrm{op}}∥⋅∥op​ is the operator norm for the Euclidean norms on complex coordinate spaces. For a matrix BBB, write λ(B)=sup⁡{r∈R:rI−B is not invertible}\lambda(B)=\sup\{r\in\mathbb R:rI-B\text{ is not invertible}\}λ(B)=sup{r∈R:rI−B is not invertible}, the supremum of its real spectrum (for the Hermitian matrices Y(ω)Y(\omega)Y(ω) this is their largest eigenvalue). The conclusions are simultaneously v=∥∑k=0N−1Ak2∥opv=\left\|\sum_{k=0}^{N-1}A_k^2\right\|_{\mathrm{op}}v=​∑k=0N−1​Ak2​​op​, ∫Ωλ(Y(ω)) dμ(ω)≤2vlog⁡d\int_\Omega\lambda(Y(\omega))\,d\mu(\omega)\le\sqrt{2v\log d}∫Ω​λ(Y(ω))dμ(ω)≤2vlogd​, and, for every real t≥0t\ge0t≥0, μ{ω:t≤λ(Y(ω))}≤T(d,v,t)\mu\{\omega:t\le\lambda(Y(\omega))\}\le T(d,v,t)μ{ω:t≤λ(Y(ω))}≤T(d,v,t), where the probability is read as a real number and T(d,v,t)=dT(d,v,t)=dT(d,v,t)=d if v=0v=0v=0 and t=0t=0t=0, T(d,v,t)=0T(d,v,t)=0T(d,v,t)=0 if v=0v=0v=0 and t≠0t\ne0t=0, and T(d,v,t)=dexp⁡(−t2/(2v))T(d,v,t)=d\exp(-t^2/(2v))T(d,v,t)=dexp(−t2/(2v)) if v≠0v\ne0v=0. Here log⁡\loglog is the natural logarithm; d=0d=0d=0 is excluded, d=1d=1d=1 is included and gives expectation upper bound 000, and N=0N=0N=0 is included, with all indexed hypotheses vacuous, Y=0Y=0Y=0, v=0v=0v=0, and the tail event having probability 111 at t=0t=0t=0 and 000 at t>0t>0t>0. The same piecewise tail convention applies whenever v=0v=0v=0, including for nonempty families, so the bound at t=0t=0t=0 is ddd and there is no division by zero in that branch. No separate integrability hypotheses are stated; integrals use the total Bochner-integral convention, which assigns zero to a nonintegrable function.

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