Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proof of Theorem 1, p. 36, last display — ∇φ Σ_{R,P} ∇φᵀ = (I − γP̃^{π*})⁻¹W⁻¹[D_R + γ²D_Q]((I − γP̃^{π*})⁻¹)ᵀ

Proved
RLInference.QValueCLT.covariance_identity

by mikedeng1 · Oct 22, 2026 · Mathlib 0df444a (Lean v4.33.1)

covariancedelta-methodmatrix-identityp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1reinforcement-learning

Let Q=Fix⁡(γ,μR,P)Q=\operatorname{Fix}(\gamma,\mu_R,P)Q=Fix(γ,μR​,P), V∗(k)=max⁡aQ(k,a)V^*(k)=\max_aQ(k,a)V∗(k)=maxa​Q(k,a), let a∗a^*a∗ be any action map and w>0w>0w>0 any positive weight vector. Let

∇ϕ=(I−γP~π∗)−1[ I  ∣  γblockdiag⁡((V∗)T,…,(V∗)T) ]\nabla\phi=(I-\gamma\tilde P^{\pi^*})^{-1}\big[\,I\;\big|\;\gamma\operatorname{blockdiag}((V^*)^T,\dots,(V^*)^T)\,\big]∇ϕ=(I−γP~π∗)−1[I​γblockdiag((V∗)T,…,(V∗)T)]

be the gradient of App. A.1, and ΣR,P=diag⁡(W−1DR,DP)\Sigma_{R,P}=\operatorname{diag}(W^{-1}D_R,D_P)ΣR,P​=diag(W−1DR​,DP​) the covariance (8). Then

∇ϕ ΣR,P ∇ϕT=(I−γP~π∗)−1W−1[DR+γ2DQ]((I−γP~π∗)−1)T,\nabla\phi\,\Sigma_{R,P}\,\nabla\phi^T=(I-\gamma\tilde P^{\pi^*})^{-1}W^{-1}\big[D_R+\gamma^2D_Q\big]\big((I-\gamma\tilde P^{\pi^*})^{-1}\big)^T,∇ϕΣR,P​∇ϕT=(I−γP~π∗)−1W−1[DR​+γ2DQ​]((I−γP~π∗)−1)T,

which is the covariance Σ\SigmaΣ of (5)–(6).

This identity turns the delta-method covariance into the closed form stated in Theorem 1.

Formalization Note. The identity is pure matrix algebra, with matrices indexed by S×A\mathcal S\times\mathcal AS×A and (S×A)×S(\mathcal S\times\mathcal A)\times\mathcal S(S×A)×S. It does not use stationarity of www, the strict argmax, or γ<1\gamma<1γ<1; only w>0w>0w>0 is assumed, as in the paper.

Preamble
import Mathlib
import Definitions.Def_RLInference_QValueCLT_Setting
open MeasureTheory ProbabilityTheory Filter Matrix Topology
Formal statement
namespace RLInference.QValueCLT

theorem covariance_identity {S A : Type} [Fintype S] [DecidableEq S] [Nonempty S]
    [Fintype A] [DecidableEq A] [Nonempty A]
    (γ : ℝ) (P : S × A → PMF S) (R : S × A → Measure ℝ) (aStar : S → A)
    (w : S × A → ℝ) (hw_pos : ∀ x, 0 < w x) :
    gradPhi γ (Pmat P) aStar (Vstar (bellmanFix γ (muR R) (Pmat P))) *
        SigmaRP (Pmat P) (sigma2R R) w *
        (gradPhi γ (Pmat P) aStar (Vstar (bellmanFix γ (muR R) (Pmat P))))ᵀ =
      SigmaQ γ (Pmat P) aStar (Vstar (bellmanFix γ (muR R) (Pmat P))) (sigma2R R) w := by sorry

end RLInference.QValueCLT
Source
Zhu, Dong & Lam, Uncertainty Quantification and Exploration for Reinforcement Learning, arXiv:1910.05471v3, p. 36, App. A.1, last display; (5)–(6) p. 10, (8) p. 12
Human review
  • Endorsed by Shuze Chen · Oct 23, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 23, 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