Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma 1 — W(t+1) = (1 − α(t))W(t) + α(t)w(t) converges to 0 w.p.1 for martingale-difference noise with bounded conditional variance

Proved
AsyncSA.Monotone.lemma1

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

martingalep2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1stochastic-approximation

Let (Ω,F,P)(\Omega,\mathcal F,P)(Ω,F,P) be a probability space and {F(t)}t≥0\{\mathcal F(t)\}_{t\ge0}{F(t)}t≥0​ an increasing sequence of σ-fields. For each ttt let α(t)\alpha(t)α(t), w(t−1)w(t-1)w(t−1) and B(t)B(t)B(t) be F(t)\mathcal F(t)F(t)-measurable scalar random variables, and let CCC be a deterministic constant. Suppose that, with probability 1,

  1. E[w(t)∣F(t)]=0E[w(t)\mid\mathcal F(t)]=0E[w(t)∣F(t)]=0 for all ttt;
  2. E[w2(t)∣F(t)]≤B(t)E[w^2(t)\mid\mathcal F(t)]\le B(t)E[w2(t)∣F(t)]≤B(t) for all ttt;
  3. α(t)∈[0,1]\alpha(t)\in[0,1]α(t)∈[0,1] for all ttt;
  4. ∑t=0∞α(t)=∞\sum_{t=0}^\infty\alpha(t)=\infty∑t=0∞​α(t)=∞;
  5. ∑t=0∞α2(t)≤C\sum_{t=0}^\infty\alpha^2(t)\le C∑t=0∞​α2(t)≤C;

and that the sequence {B(t)}\{B(t)\}{B(t)} is bounded with probability 1. Let W(t)W(t)W(t) satisfy

W(t+1)=(1−α(t))W(t)+α(t)w(t),t≥0,W(t+1)=(1-\alpha(t))W(t)+\alpha(t)w(t),\qquad t\ge0,W(t+1)=(1−α(t))W(t)+α(t)w(t),t≥0,

with W(0)W(0)W(0) arbitrary. Then

lim⁡t→∞W(t)=0with probability 1.\lim_{t\to\infty}W(t)=0\quad\text{with probability 1.}t→∞lim​W(t)=0with probability 1.

This is the scalar stochastic approximation result on which the convergence analysis of the asynchronous algorithm rests: in §5 it is applied, component by component, to the accumulated noise Wi(t)W_i(t)Wi​(t).

Formalization Note The conditional expectations are generalized: (1) means ∫Sw(t) dP=0\int_S w(t)\,dP=0∫S​w(t)dP=0 for every S∈F(t)S\in\mathcal F(t)S∈F(t) on which w(t)w(t)w(t) is integrable, and (2) means ∫Sw(t)2 dP≤∫SB(t)+ dP\int_S w(t)^2\,dP\le\int_S B(t)^+\,dP∫S​w(t)2dP≤∫S​B(t)+dP in [0,∞][0,\infty][0,∞] for every S∈F(t)S\in\mathcal F(t)S∈F(t); no integrability of w(t)w(t)w(t) is assumed, as in the paper. (4) is divergence of the partial sums to +∞+\infty+∞ and (5) bounds every partial sum by CCC. "w(t−1)w(t-1)w(t−1) is F(t)\mathcal F(t)F(t)-measurable" is written as "w(t)w(t)w(t) is F(t+1)\mathcal F(t+1)F(t+1)-measurable". The boundedness of {B(t)}\{B(t)\}{B(t)} is sup⁡t∣B(t)∣<∞\sup_t|B(t)|<\inftysupt​∣B(t)∣<∞ on almost every sample path, with a bound that may depend on the path. The recursion holds on every sample path, and W(0)W(0)W(0) is an arbitrary (not necessarily measurable) function.

Preamble
import Mathlib
import Definitions.Def_AsyncSA_Monotone_Model
Formal statement
namespace AsyncSA.Monotone

open MeasureTheory Filter Topology

theorem lemma1 {Ω : Type*} [m0 : MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
    (𝓕 : Filtration ℕ m0) (α w B W : ℕ → Ω → ℝ) (C : ℝ)
    (hα_meas : ∀ t, Measurable[𝓕 t] (α t))
    (hw_meas : ∀ t, Measurable[𝓕 (t + 1)] (w t))
    (hB_meas : ∀ t, Measurable[𝓕 t] (B t))
    (ha : ∀ t, CondMeanZero (𝓕 t) P (w t))
    (hb : ∀ t, CondSqLe (𝓕 t) P (w t) (B t))
    (hc : ∀ᵐ ω ∂P, ∀ t, α t ω ∈ Set.Icc (0 : ℝ) 1)
    (hd : ∀ᵐ ω ∂P, Tendsto (fun T => ∑ t ∈ Finset.range T, α t ω) atTop atTop)
    (he : ∀ᵐ ω ∂P, ∀ T, ∑ t ∈ Finset.range T, α t ω ^ 2 ≤ C)
    (hB_bdd : ∀ᵐ ω ∂P, ∃ K : ℝ, ∀ t, |B t ω| ≤ K)
    (hW : ∀ t ω, W (t + 1) ω = (1 - α t ω) * W t ω + α t ω * w t ω) :
    ∀ᵐ ω ∂P, Tendsto (fun t => W t ω) atTop (𝓝 0) := by sorry

end AsyncSA.Monotone
Source
Tsitsiklis, Asynchronous Stochastic Approximation and Q-Learning, Machine Learning 16 (1994), p. 190, §3, Lemma 1
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