Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

(3.24), p. 13 — ∫_C G(x, dy)(∫_C ω_ε(y, z)v₀(z)K(y, dz) + ∫_{Cᶜ} ω̃_ε(y, z)|f(z)|K(y, dz)) < ∞

Proved
LyapDiffMC.RandomHorizon.eq_3_24

by mikedeng1 · Oct 24, 2026 · Mathlib 0df444a (Lean v4.33.1)

lyapunov-functionp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-strengthenedp2o-v1potential-kernelsensitivity-analysis

Throughout, SSS is a measurable state space, (P(θ):θ∈Λ)(P(\theta):\theta\in\Lambda)(P(θ):θ∈Λ) a family of Markov transition kernels on SSS indexed by an open interval Λ=(a,b)\Lambda=(a,b)Λ=(a,b), f,g:S→Rf, g : S\to\mathbb Rf,g:S→R measurable, and ∅≠C⊆S\emptyset\neq C\subseteq S∅=C⊆S measurable. The non-negative kernel K(θ,x,dy)=eg(x)P(θ,x,dy)K(\theta,x,dy)=e^{g(x)}P(\theta,x,dy)K(θ,x,dy)=eg(x)P(θ,x,dy) restricted to y∈Cy\in Cy∈C is K(θ)K(\theta)K(θ), its powers are Kn(θ)K^n(\theta)Kn(θ) (K0(x,dy)=δx(dy)K^0(x,dy)=\delta_x(dy)K0(x,dy)=δx​(dy)), G(θ,x,dy)=∑n≥0Kn(θ,x,dy)G(\theta,x,dy)=\sum_{n\ge0}K^n(\theta,x,dy)G(θ,x,dy)=∑n≥0​Kn(θ,x,dy) (possibly an infinite measure), f~(θ,x)=f(x)+∫CcK(θ,x,dy)f(y)\tilde f(\theta,x)=f(x)+\int_{C^c}K(\theta,x,dy)f(y)f~​(θ,x)=f(x)+∫Cc​K(θ,x,dy)f(y), and u∗(θ,x)=∑n≥0(Kn(θ)f~(θ))(x)=∫f~(θ,y) G(θ,x,dy)u^*(\theta,x)=\sum_{n\ge0}(K^n(\theta)\tilde f(\theta))(x)=\int \tilde f(\theta,y)\,G(\theta,x,dy)u∗(θ,x)=∑n≥0​(Kn(θ)f~​(θ))(x)=∫f~​(θ,y)G(θ,x,dy). Write K=K(θ0)K=K(\theta_0)K=K(θ0​) and G=G(θ0)G=G(\theta_0)G=G(θ0​).

Let ε>0\varepsilon>0ε>0, and let k(θ,x,y)k(\theta,x,y)k(θ,x,y) be measurable in (x,y)(x,y)(x,y) for θ∈[θ0−ε,θ0+ε]\theta\in[\theta_0-\varepsilon,\theta_0+\varepsilon]θ∈[θ0​−ε,θ0​+ε], with k(⋅,x,y)k(\cdot,x,y)k(⋅,x,y) continuously differentiable on [θ0−ε,θ0+ε][\theta_0-\varepsilon,\theta_0+\varepsilon][θ0​−ε,θ0​+ε] for x∈Cx\in Cx∈C and every yyy, with derivative k′k'k′. Put ωε(x,y)=sup⁡∣θ−θ0∣<ε∣k′(θ,x,y)∣\omega_\varepsilon(x,y)=\sup_{|\theta-\theta_0|<\varepsilon}|k'(\theta,x,y)|ωε​(x,y)=sup∣θ−θ0​∣<ε​∣k′(θ,x,y)∣ (ω~ε\tilde\omega_\varepsilonω~ε​ for y∈Ccy\in C^cy∈Cc) and r~ε(x)=∫Ccω~ε(x,y)∣f(y)∣K(x,dy)\tilde r_\varepsilon(x)=\int_{C^c}\tilde\omega_\varepsilon(x,y)|f(y)|K(x,dy)r~ε​(x)=∫Cc​ω~ε​(x,y)∣f(y)∣K(x,dy). Let v0,v1v_0,v_1v0​,v1​ be finite, non-negative, measurable functions on CCC satisfying the Lyapunov bound

(Kv1)(x)≤v1(x)−∫Cωε(x,y)v0(y)K(x,dy)−r~ε(x)(x∈C).(3.12)(Kv_1)(x)\le v_1(x)-\int_C\omega_\varepsilon(x,y)v_0(y)K(x,dy)-\tilde r_\varepsilon(x)\quad(x\in C).\tag{3.12}(Kv1​)(x)≤v1​(x)−∫C​ωε​(x,y)v0​(y)K(x,dy)−r~ε​(x)(x∈C).(3.12)

Then for every x∈Cx\in Cx∈C,

∫CG(x,dy)(∫Cωε(y,z)v0(z)K(y,dz)+∫Ccω~ε(y,z)∣f(z)∣K(y,dz))<∞.(3.24)\int_C G(x,dy)\Big(\int_C\omega_\varepsilon(y,z)v_0(z)K(y,dz)+\int_{C^c}\tilde\omega_\varepsilon(y,z)|f(z)|K(y,dz)\Big)<\infty.\tag{3.24}∫C​G(x,dy)(∫C​ωε​(y,z)v0​(z)K(y,dz)+∫Cc​ω~ε​(y,z)∣f(z)∣K(y,dz))<∞.(3.24)

Together with the pointwise bounds (3.22)–(3.23), this is the integrable majorant for the dominated-convergence argument that proves differentiability of u∗u^*u∗ at θ0\theta_0θ0​.

Formalization Note The Lyapunov inequalities are written additively in [0,∞][0,\infty][0,∞] (Kv+⋯≤vKv+\dots\le vKv+⋯≤v), which is equivalent because v0,v1v_0,v_1v0​,v1​ are finite; ωε\omega_\varepsilonωε​ and r~ε\tilde r_\varepsilonr~ε​ are [0,∞][0,\infty][0,∞]-valued. k′k'k′ is the derivative within [θ0−ε,θ0+ε][\theta_0-\varepsilon,\theta_0+\varepsilon][θ0​−ε,θ0​+ε], the ordinary derivative at interior points. One ε\varepsilonε serves A2, A3 and the Lyapunov conditions, as in the paper's proof. The family P(θ)P(\theta)P(θ) is given for all real θ\thetaθ; only θ∈Λ\theta\in\Lambdaθ∈Λ matters.

Differences from the source:

  • generalization: the hypotheses of Theorem 3.1 are in force → only (3.12), A2 (k(·, x, y) C¹ on [θ₀ − ε, θ₀ + ε]), measurability of k(θ, ·, ·) on that interval and measurable v₀, v₁ ≥ 0 on C; reason: (3.24) follows from (3.12) and Proposition 3.1 (p. 13); A2 and the measurability of k make ω_ε measurable; under the theorem's hypotheses (with [θ₀ − ε, θ₀ + ε] ⊆ Λ) the statement specializes to the paper's
  • encoding: separate ε's in A2, A3 and the theorem → one ε; reason: the proof takes the smallest of them (p. 12); shrinking ε keeps every hypothesis true, since ω_ε and r̃_ε only decrease
  • encoding: (K v)(x) ≤ v(x) − (non-negative terms) in (3.11), (3.12), (3.12′) → (K v)(x) + (the terms) ≤ v(x) in ℝ≥0∞; reason: every term lies in [0, ∞] and v(x) is finite, so the two forms are equivalent
  • encoding: v₀, v₁ finite-valued non-negative functions on C → v₀ v₁ : S → ℝ, non-negative on C; reason: values off C never enter the statement; same instances
  • encoding: ω_ε(x, y) = sup{|k′(θ, x, y)| : |θ − θ₀| < ε}, ω̃_ε, r̃_ε → ℝ≥0∞-valued iSup (omega) and lintegral (rtilde); reason: the supremum may be +∞; a real supremum would return 0 on an unbounded set
  • encoding: k′(θ, x, y), the derivative of k(·, x, y) on [θ₀ − ε, θ₀ + ε] → kder = derivWithin on Icc (θ₀ − ε) (θ₀ + ε); reason: the ordinary derivative at interior points (in particular at θ₀), the one-sided one at the endpoints, as A2's 'in [θ₀ − ε, θ₀ + ε]'
  • encoding: measurability of v₀, v₁ left implicit → Measurable v₀, Measurable v₁; reason: (Kv)(x) is only defined for measurable v
Preamble
import Mathlib
import Definitions.Def_LyapDiffMC_RandomHorizon_Model

open MeasureTheory ProbabilityTheory
open scoped ENNReal
Formal statement
namespace LyapDiffMC.RandomHorizon

/-- (3.24), p. 13. The Lyapunov bound (3.12) together with Proposition 3.1 gives
`∫_C G(x, dy)(∫_C ω_ε(y, z) v₀(z) K(y, dz) + ∫_{Cᶜ} ω̃_ε(y, z)|f(z)| K(y, dz)) < ∞` for `x ∈ C`. -/
theorem eq_3_24 {S : Type*} [MeasurableSpace S] (M : Setting S)
    (θ₀ ε : ℝ) (hε : 0 < ε) (k : ℝ → S → S → ℝ)
    (hk_meas : ∀ θ ∈ Set.Icc (θ₀ - ε) (θ₀ + ε), Measurable (Function.uncurry (k θ)))
    (hA2 : ∀ x ∈ M.C, ∀ y, ContDiffOn ℝ 1 (fun θ => k θ x y) (Set.Icc (θ₀ - ε) (θ₀ + ε)))
    (v₀ v₁ : S → ℝ) (hv₀ : Measurable v₀) (hv₁ : Measurable v₁)
    (hv₀_nn : ∀ x ∈ M.C, 0 ≤ v₀ x) (hv₁_nn : ∀ x ∈ M.C, 0 ≤ v₁ x)
    (h312 : ∀ x ∈ M.C,
      ∫⁻ y in M.C, ENNReal.ofReal (v₁ y) ∂(discKernel M θ₀ x)
        + ∫⁻ y in M.C, omega k θ₀ ε x y * ENNReal.ofReal (v₀ y) ∂(discKernel M θ₀ x)
        + rtilde M k θ₀ ε x ≤ ENNReal.ofReal (v₁ x)) :
    ∀ x ∈ M.C,
      ∫⁻ y, (∫⁻ z in M.C, omega k θ₀ ε y z * ENNReal.ofReal (v₀ z) ∂(discKernel M θ₀ y)
          + ∫⁻ z in M.Cᶜ, omega k θ₀ ε y z * ‖M.f z‖ₑ ∂(discKernel M θ₀ y))
        ∂(greenMeasure M θ₀ x) < ⊤ := by sorry

end LyapDiffMC.RandomHorizon
Source
Rhee, Glynn, Lyapunov Conditions for Differentiability of Markov Chain Expectations, arXiv:1707.03870v1, p. 13, (3.24)
Human review
  • Endorsed by Shuze Chen · Oct 24, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 24, 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