Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proposition 4.2 [Grincevičius (1980)] — Lévy-type maximal inequality for the partial sums of a perpetuity

Proved
GoldieRenewal.Kesten.proposition_4_2

by mikedeng1 · Oct 6, 2026 · Mathlib 0df444a (Lean v4.33.1)

maximal-inequalityp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1perpetuityrandom-difference-equation

Let (Qn,Mn)(Q_n,M_n)(Qn​,Mn​), n=1,2,…n=1,2,\dotsn=1,2,…, be independent pairs of real random variables, each with the law μ\muμ of (Q,M)(Q,M)(Q,M), and let Πj\Pi_jΠj​, RnR_nRn​, Πj,n\Pi_{j,n}Πj,n​, Rj,nR_{j,n}Rj,n​ be the partial products and sums

Πj=∏k=1jMk,Rn=∑k=1nΠk−1Qk,Πj,n=∏k=j+1nMk,Rj,n=∑k=j+1nΠj,k−1Qk,\Pi_j=\prod_{k=1}^j M_k,\quad R_n=\sum_{k=1}^n\Pi_{k-1}Q_k,\quad \Pi_{j,n}=\prod_{k=j+1}^n M_k,\quad R_{j,n}=\sum_{k=j+1}^n\Pi_{j,k-1}Q_k ,Πj​=k=1∏j​Mk​,Rn​=k=1∑n​Πk−1​Qk​,Πj,n​=k=j+1∏n​Mk​,Rj,n​=k=j+1∑n​Πj,k−1​Qk​,

so that Rn=Rj+ΠjRj,nR_n=R_j+\Pi_jR_{j,n}Rn​=Rj​+Πj​Rj,n​. Then for every n≥1n\ge 1n≥1 and all x,y∈Rx,y\in\mathbb Rx,y∈R,

P(max⁡j=1,…,n(Rj+Πj med(Rj,n+Πj,ny))>x)≤2P(Rn+Πny>x),P\Big(\max_{j=1,\dots,n}\big(R_j+\Pi_j\,\mathrm{med}(R_{j,n}+\Pi_{j,n}y)\big)>x\Big)\le 2P\big(R_n+\Pi_ny>x\big),P(j=1,…,nmax​(Rj​+Πj​med(Rj,n​+Πj,n​y))>x)≤2P(Rn​+Πn​y>x),

where med(X)\mathrm{med}(X)med(X) is a median of the law of XXX, and the inequality holds for every choice of medians.

This extends Lévy's symmetrization inequality from sums of independent variables to the affine recursion; in the proof of Theorem 4.1 it bounds the probability that some partial sum is large by the tail of the limit.

Formalization Note The sequence is stored 0-based in Lean (the iii-th Lean variable is the paper's (Qi+1,Mi+1)(Q_{i+1},M_{i+1})(Qi+1​,Mi+1​)). "The maximum exceeds xxx" is written as "some j∈{1,…,n}j\in\{1,\dots,n\}j∈{1,…,n} has the term exceeding xxx". No condition on (Q,M)(Q,M)(Q,M) beyond measurability is assumed, as in the paper, which notes that the assumption P(M=0)=0P(M=0)=0P(M=0)=0 of Grincevičius (1981) is not needed.

Preamble
import Mathlib
import Definitions.Def_GoldieRenewal_Kesten_Perpetuity
Formal statement
namespace GoldieRenewal.Kesten

open MeasureTheory ProbabilityTheory
open scoped ENNReal

/-- **Proposition 4.2** [Grincevičius (1980)] (Goldie, *Implicit renewal theory and tails of
solutions of random equations*, Ann. Appl. Probab. 1(1) (1991), p. 136). Let `(Q_n, M_n)`,
`n = 1, 2, …`, be independent, each with the law `μ` of `(Q, M)`. With
`Π_j = M₁ ⋯ M_j`, `R_n = Σ_{k=1}^n Π_{k−1} Q_k`, `Π_{j,n} = Π_{k=j+1}^n M_k`,
`R_{j,n} = Σ_{k=j+1}^n Π_{j,k−1} Q_k`, for all `x, y ∈ ℝ`,
`P(max_{j=1,…,n} (R_j + Π_j med(R_{j,n} + Π_{j,n} y)) > x) ≤ 2 P(R_n + Π_n y > x)`.

**Formalization Note** The sequence is stored 0-based: `Q i`, `M i` (`i = 0, 1, …`) are the
paper's `Q_{i+1}`, `M_{i+1}` (see `piProd`, `partialSum`, `piProdFrom`, `partialSumFrom`).
`med X` is any median of the law of `X`; the paper fixes none, so the inequality is asserted for
every choice `med j` of medians, `j = 1, …, n`. "max > x" is "some `j ∈ {1, …, n}` has `… > x`".
No moment or Cramér condition is assumed, as in the paper. -/
theorem proposition_4_2 {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
    (μ : Measure (ℝ × ℝ)) (Q M : ℕ → Ω → ℝ) (hQ : ∀ i, Measurable (Q i))
    (hM : ∀ i, Measurable (M i))
    (hindep : iIndepFun (fun i ω => (Q i ω, M i ω)) P)
    (hlaw : ∀ i, P.map (fun ω => (Q i ω, M i ω)) = μ)
    (n : ℕ) (x y : ℝ) (med : ℕ → ℝ)
    (hmed : ∀ j ∈ Finset.Icc 1 n,
      IsMedian (P.map (fun ω => partialSumFrom Q M j n ω + piProdFrom M j n ω * y)) (med j)) :
    P {ω | ∃ j ∈ Finset.Icc 1 n, x < partialSum Q M j ω + piProd M j ω * med j}
      ≤ 2 * P {ω | x < partialSum Q M n ω + piProd M n ω * y} := by sorry

end GoldieRenewal.Kesten
Source
Goldie, Implicit renewal theory and tails of solutions of random equations, Ann. Appl. Probab. 1(1):126–166 (1991), DOI 10.1214/aoap/1177005985, p. 136, Proposition 4.2 [Grincevičius (1980)]
Human review
  • Endorsed by Shuze Chen · Oct 6, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 6, 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