Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 3.6 (Master Tail Bound) — P{λmax(Σ X_k) ≥ t} ≤ e^{−θt}·tr exp(Σ_k log E e^{θX_k})

Proved
MatrixTail.Master.theorem3_6

by mikedeng1 · Oct 7, 2026 · Mathlib 0df444a (Lean v4.33.1)

master-inequalityp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1random-matricestail-bound

Let X1,…,XnX_1,\dots,X_nX1​,…,Xn​ be independent random d×dd\times dd×d complex Hermitian matrices on a probability space, with d≥1d\ge1d≥1. For every t∈Rt\in\mathbb Rt∈R and every θ>0\theta>0θ>0 at which all the matrix moment generating functions EeθXk\mathbb E e^{\theta X_k}EeθXk​ exist,

P{λmax⁡(∑kXk)≥t}≤e−θt⋅tr⁡exp⁡(∑klog⁡EeθXk).\mathbb P\Bigl\{\lambda_{\max}\Bigl(\sum_k X_k\Bigr)\ge t\Bigr\} \le e^{-\theta t}\cdot\operatorname{tr}\exp\Bigl(\sum_k\log\mathbb E e^{\theta X_k}\Bigr).P{λmax​(k∑​Xk​)≥t}≤e−θt⋅trexp(k∑​logEeθXk​).

Equivalently, the left side is at most inf⁡θ>0{e−θttr⁡exp⁡(∑klog⁡EeθXk)}\inf_{\theta>0}\bigl\{e^{-\theta t}\operatorname{tr}\exp\bigl(\sum_k\log\mathbb E e^{\theta X_k}\bigr)\bigr\}infθ>0​{e−θttrexp(∑k​logEeθXk​)}, which is display (3.5) of the paper.

This master inequality is the progenitor of the matrix Gaussian-series, Chernoff, Bernstein and Azuma bounds of the paper: each follows by bounding the individual matrix cumulant generating functions in the semidefinite order.

Formalization Note. The infimum over θ>0\theta>0θ>0 is stated for every θ>0\theta>0θ>0 at which the moment generating functions exist (every entry of eθXke^{\theta X_k}eθXk​ integrable); the paper admits non-existent ones, for which the right side is +∞+\infty+∞, so the reading is exact. The dimension is positive, since for d=0d=0d=0 the largest eigenvalue would be 000 and the bound would fail for t≤0t\le0t≤0.

Preamble
import Mathlib
import Definitions.Def_MatrixTail_Master_Model

open MeasureTheory ProbabilityTheory
open scoped MatrixOrder ComplexOrder ENNReal
Formal statement
namespace MatrixTail.Master

/-- **Theorem 3.6 (Master Tail Bound for Independent Sums).** Tropp, *User-Friendly Tail Bounds for Sums of
Random Matrices*, arXiv:1004.4389v7, p. 12, display (3.5): "Consider a finite sequence `{X_k}` of independent,
random, self-adjoint matrices. For all `t ∈ ℝ`,
`P{λmax(Σ_k X_k) ≥ t} ≤ inf_{θ>0} { e^{−θt} · tr exp(Σ_k log E e^{θX_k}) }`."

Proof on the page: "Substitute the subadditivity rule for matrix cgfs, Lemma 3.4, into the Laplace transform
bound, Proposition 3.1."

**Formalization Note.**
* The finite sequence is `X : Fin n → Ω → Matrix (Fin d) (Fin d) ℂ` (complex matrices, §2.1), each `X k`
  measurable and Hermitian at every outcome, jointly independent (`iIndepFun X P`) under a probability
  measure `P`; the dimension is `d ≥ 1` (`[NeZero d]`), since for `d = 0`, `λmax = sSup ∅ = 0` and the
  bound would fail at `t ≤ 0`.
* The infimum over `θ > 0` is stated as the bound for every `θ > 0` at which all the mgfs `E e^{θX_k}` exist
  (every entry of `e^{θX_k}` integrable, `MatIntegrable`: the §2.2 regularity made explicit). The paper admits
  non-existent mgfs (p. 9), for which the right side is `+∞`; since `P ≤ inf_θ F(θ)` iff `P ≤ F(θ)` for
  every `θ`, this reading is exact. No real-valued `⨅` is used (it would be `0` on a set unbounded below).
* `E` is the entrywise expectation `mean`; `exp`/`log` are `cfc Real.exp`/`cfc Real.log`; `tr exp(·)` is
  `trExp`; `λmax` is `lambdaMax`. -/
theorem theorem3_6 {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
    {d n : ℕ} [NeZero d] (X : Fin n → Ω → Matrix (Fin d) (Fin d) ℂ)
    (hX_meas : ∀ k, Measurable (X k)) (hX_herm : ∀ k ω, (X k ω).IsHermitian)
    (hX_indep : iIndepFun X P) :
    ∀ (t θ : ℝ), 0 < θ → (∀ k, MatIntegrable P (fun ω => mexp (θ • X k ω))) →
      P.real {ω | t ≤ lambdaMax (∑ k, X k ω)} ≤
        Real.exp (-θ * t) * trExp (∑ k, mlog (mean P (fun ω => mexp (θ • X k ω)))) := by sorry

end MatrixTail.Master
Source
Tropp, User-Friendly Tail Bounds for Sums of Random Matrices, arXiv:1004.4389v7, p. 12, Theorem 3.6, display (3.5)
Human review
  • Endorsed by Shuze Chen · Oct 8, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 8, 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