Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Lemma 6 — the SA scheme reaches E[∥zi,t−x^i(yk)∥2∣Fk]≤Qi/(t+1)\mathbb E[\|z_{i,t} - \widehat x_i(y_k)\|^2 \mid \mathcal F_k] \le Q_i/(t+1)E[∥zi,t​−xi​(yk​)∥2∣Fk​]≤Qi​/(t+1)

Proved
LeiBR.Rand.lemma_6

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

conditional-expectationp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1projected-gradientstochastic-approximation

Let Assumption 1 hold and μ>0\mu > 0μ>0. Let yky_kyk​ be an Fk\mathcal F_kFk​-measurable random profile with values in XXX. Let player iii's samples ξi,kt\xi^t_{i,k}ξi,kt​, t≥1t \ge 1t≥1, be i.i.d. with the law of ξ\xiξ, and let the vector ξi,k=(ξi,k1,ξi,k2,… )\xi_{i,k} = (\xi^1_{i,k}, \xi^2_{i,k}, \dots)ξi,k​=(ξi,k1​,ξi,k2​,…) be independent of Fk\mathcal F_kFk​. Run the projected stochastic gradient scheme (SAi,k_{i,k}i,k​) from zi,1=yi,kz_{i,1} = y_{i,k}zi,1​=yi,k​, with γt=1/(μ(t+1))\gamma_t = 1/(\mu(t+1))γt​=1/(μ(t+1)). Then for every t≥1t \ge 1t≥1,

E[∥zi,t−x^i(yk)∥2 ∣ Fk]≤Qit+1a.s.,\mathbb E\big[\|z_{i,t} - \widehat x_i(y_k)\|^2 \,\big|\, \mathcal F_k\big] \le \frac{Q_i}{t+1} \quad \text{a.s.},E[∥zi,t​−xi​(yk​)∥2​Fk​]≤t+1Qi​​a.s.,

with Qi=2Mi2/μ2+2DXi2Q_i = 2M_i^2/\mu^2 + 2D_{X_i}^2Qi​=2Mi2​/μ2+2DXi​2​.

It is the inner-loop guarantee: jjj steps of (SAi,k_{i,k}i,k​) produce an inexact best response of accuracy Qi/jQ_i/jQi​/j in conditional mean square, which is how the inexactness condition (33) is met.

Formalization Note

  • The paper states the lemma for i∈Iki \in I_ki∈Ik​, the players with χi,k=1\chi_{i,k} = 1χi,k​=1. Lean states it for any player, since the samples are drawn regardless of the coin.
  • In Algorithm 2, zi,1=xi,kz_{i,1} = x_{i,k}zi,1​=xi,k​ and yk=xky_k = x_kyk​=xk​.
  • ΠXi\Pi_{X_i}ΠXi​​ is any map satisfying SpectralProjGrad.Shared.IsProjOnto.
Preamble
import Mathlib
import Definitions.Def_SpectralProjGrad_Shared_IsProjOnto
import Definitions.Def_LeiBR_Rand_Game
import Definitions.Def_LeiBR_Rand_ProxBR
import Definitions.Def_LeiBR_Rand_Assumption1
import Definitions.Def_LeiBR_Rand_SA
import Definitions.Def_LeiBR_Rand_Algorithm2
import Definitions.Def_LeiBR_Rand_Constants

open MeasureTheory ProbabilityTheory
Formal statement
namespace LeiBR.Rand

/-- Lemma 6, p. 15. Let `y_k` be an `F_k`-measurable `X`-valued random profile and let player `i`'s
samples `ξ^t` (`t ≥ 1`) be i.i.d. with the law `μξ` of `ξ` and, as a vector, independent of `F_k`.
Run (SA_{i,k}) from `z_{i,1} = y_{i,k}`. Then for every `t ≥ 1`,
`E[‖z_{i,t} − x̂_i(y_k)‖² | F_k] ≤ Q_i/(t+1)` a.s. -/
theorem lemma_6 {N : ℕ} {n : Fin N → ℕ} {d : ℕ} (G : Game N n)
    (μξ : Measure (EuclideanSpace ℝ (Fin d)))
    (ψ : Fin N → LeiBR.Sync.Profile n → EuclideanSpace ℝ (Fin d) → ℝ)
    (gψ : ∀ i, LeiBR.Sync.Profile n → EuclideanSpace ℝ (Fin d) → LeiBR.Sync.Strat n i) (M : Fin N → ℝ)
    (hA1 : Assumption1 G μξ ψ gψ M) (mu : ℝ) (hmu : 0 < mu)
    (xhat : LeiBR.Sync.Profile n → LeiBR.Sync.Profile n) (hxhat : IsProxBR G mu xhat)
    (proj : ∀ i, LeiBR.Sync.Strat n i → LeiBR.Sync.Strat n i)
    (hproj : ∀ i, SpectralProjGrad.Shared.IsProjOnto (G.X i) (proj i))
    {Ω : Type*} [mΩ : MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
    (F : Filtration ℕ mΩ) (k : ℕ)
    (y : Ω → LeiBR.Sync.Profile n) (hy_meas : StronglyMeasurable[F k] y) (hy_feas : ∀ ω, G.Feasible (y ω))
    (s : ℕ → Ω → EuclideanSpace ℝ (Fin d)) (hs_meas : ∀ t, Measurable (s t))
    (hs_law : ∀ t, Measure.map (s (t + 1)) P = μξ) (hs_iid : iIndepFun (fun t => s (t + 1)) P)
    (hs_indep : Indep (⨆ t, MeasurableSpace.comap (s (t + 1)) inferInstance) (F k) P)
    (i : Fin N) (t : ℕ) (ht : 1 ≤ t) :
    P[fun ω => ‖saIter (proj i) (fun t' z => saDir gψ mu i (y ω) (s t' ω) z) mu (y ω i) t
        - xhat (y ω) i‖ ^ 2 | F k] ≤ᵐ[P] fun _ => Qconst G M mu i / ((t : ℝ) + 1) := by sorry

end LeiBR.Rand
Source
Lei, Shanbhag, Pang & Sen, On Synchronous, Asynchronous, and Randomized Best-Response Schemes for Stochastic Nash Games, arXiv:1704.04578v2, p. 15, Lemma 6 (cf. Lemma 3, p. 10)
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