Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 2.6 — asymptotic distributions of solutions of a generalized equation with random data

Proved
KingRockAsymp.Distribution.theorem_2_6

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

asymptotic-statisticsconvergence-in-distributiongeneralized-equationsp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1

Let ZZZ be a separable Banach space with its Borel σ\sigmaσ-algebra, and assume the analytical assumptions M.1–M.4 at (z∗,x∗)(z^*,x^*)(z∗,x∗) for f:Z×Rn→Rmf : Z \times \mathbb R^n \to \mathbb R^mf:Z×Rn→Rm and N:Rn⇉RmN : \mathbb R^n \rightrightarrows \mathbb R^mN:Rn⇉Rm, with strong partial B-derivative Dzf(z∗,x∗)D_z f(z^*,x^*)Dz​f(z∗,x∗); let F=f(z∗,⋅)+NF = f(z^*,\cdot) + NF=f(z∗,⋅)+N. Let τν>0\tau_\nu > 0τν​>0 be numbers with τν→0\tau_\nu \to 0τν​→0, and let zνz^\nuzν be random variables in ZZZ on a probability space (Ω,P)(\Omega,P)(Ω,P) such that

τν−1[zν−z∗]→ D w\tau_\nu^{-1}\big[z^\nu - z^*\big] \xrightarrow{\ \mathcal D\ } wτν−1​[zν−z∗] D ​w

for a random variable www in ZZZ (on its own probability space). Let xνx^\nuxν be measurable maps such that, for every ν\nuν, xνx^\nuxν is almost surely a solution of 0∈f(zν,x)+N(x)0 \in f(z^\nu, x) + N(x)0∈f(zν,x)+N(x), and suppose xν→x∗x^\nu \to x^*xν→x∗ almost surely. Then there is a map L:Z→RnL : Z \to \mathbb R^nL:Z→Rn with L(w)∈DF−1(0∣x∗)(−Dzf(z∗,x∗)(w))L(w) \in DF^{-1}(0|x^*)\big(-D_z f(z^*,x^*)(w)\big)L(w)∈DF−1(0∣x∗)(−Dz​f(z∗,x∗)(w)) for every w∈Zw \in Zw∈Z — that is, L(w)L(w)L(w) is the unique element of that set, by M.4 — such that

τν−1[xν−x∗]→ D L(w)=DF−1(0∣x∗)(−Dzf(z∗,x∗)(w)).\tau_\nu^{-1}\big[x^\nu - x^*\big] \xrightarrow{\ \mathcal D\ } L(w) = DF^{-1}(0|x^*)\big(-D_z f(z^*,x^*)(w)\big).τν−1​[xν−x∗] D ​L(w)=DF−1(0∣x∗)(−Dz​f(z∗,x∗)(w)).

This is the paper's abstract limit theorem: the asymptotic law of the solutions is the image of the asymptotic law of the data under the contingent derivative of F−1F^{-1}F−1, a positively homogeneous but in general nonlinear map, so the limit need not be normal even when www is.

Formalization Note The paper states: "if a sequence {xν}\{x^\nu\}{xν} of measurable selections … converges almost surely, it converges to the point x∗x^*x∗". This fails when J(z∗)J(z^*)J(z∗) has another point: with n=m=1n = m = 1n=m=1, Z=RZ = \mathbb RZ=R, N≡{0}N \equiv \{0\}N≡{0}, f(z,x)=x2−x−zf(z,x) = x^2 - x - zf(z,x)=x2−x−z, z∗=x∗=0z^* = x^* = 0z∗=x∗=0, assumptions M.1–M.4 hold, J(0)={0,1}J(0) = \{0,1\}J(0)={0,1}, and solutions near 111 converge almost surely to 111. We therefore assume that the almost sure limit is x∗x^*x∗, and drop that clause from the conclusion. "{τν}\{\tau_\nu\}{τν​} tends to 000" is from p. 7. Convergence in distribution is Mathlib's TendstoInDistribution (weak convergence of laws), with the limit on a separate probability space. The selection property is required almost surely for every ν\nuν.

Preamble
import Mathlib
import Definitions.Def_KingRockAsymp_Distribution_Basic
Formal statement
namespace KingRockAsymp.Distribution

open Set Filter Topology Metric MeasureTheory ProbabilityTheory

theorem theorem_2_6 {Z : Type*} [NormedAddCommGroup Z] [NormedSpace ℝ Z] [CompleteSpace Z]
    [TopologicalSpace.SeparableSpace Z] [MeasurableSpace Z] [BorelSpace Z] {n m : ℕ}
    {Ω Ω' : Type*} [MeasurableSpace Ω] [MeasurableSpace Ω']
    (P : Measure Ω) [IsProbabilityMeasure P] (μ' : Measure Ω') [IsProbabilityMeasure μ']
    (f : Z → Rn n → Rn m) (N : Rn n → Set (Rn m)) (z₀ : Z) (x₀ : Rn n) (Dz : Z → Rn m)
    (hM : AnalyticalAssumptions f N z₀ x₀ Dz)
    (τ : ℕ → ℝ) (hτpos : ∀ ν, 0 < τ ν) (hτ : Tendsto τ atTop (𝓝 0))
    (z : ℕ → Ω → Z) (hz : ∀ ν, Measurable (z ν))
    (W : Ω' → Z) (hW : Measurable W)
    (hconv : TendstoInDistribution (fun ν ω => (τ ν)⁻¹ • (z ν ω - z₀)) atTop W
      (fun _ => P) μ')
    (x : ℕ → Ω → Rn n) (hx : ∀ ν, Measurable (x ν))
    (hsel : ∀ ν, ∀ᵐ ω ∂P, x ν ω ∈ solMap f N (z ν ω))
    (hlim : ∀ᵐ ω ∂P, Tendsto (fun ν => x ν ω) atTop (𝓝 x₀)) :
    ∃ L : Z → Rn n,
      (∀ w : Z, L w ∈ contingentDeriv (svInv (fPlusN f N z₀)) 0 x₀ (-(Dz w))) ∧
      TendstoInDistribution (fun ν ω => (τ ν)⁻¹ • (x ν ω - x₀)) atTop
        (fun ω' => L (W ω')) (fun _ => P) μ' := by sorry

end KingRockAsymp.Distribution
Source
King & Rockafellar, Asymptotic Theory for Solutions in Statistical Estimation and Stochastic Programming, Math. Oper. Res. 18(1) (1993), Theorem 2.6, p. 8 (authors' manuscript pagination)
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