Theorem 10 — M cumulative processes on the same regeneration points are jointly asymptotically normal with covariances a_kl, and b_kl if μ₂ < ∞
ProvedSmithRegenerative.CLT.theorem_10Let be cumulative processes based on the same renewal process (, ), so that the cycle vectors are independent and identically distributed. Let be the number of regenerations in . Suppose and for every , and put
Then:
- as the random vector with components
converges in distribution to the centred normal law with covariance matrix ; 2. if in addition , the random vector with components
converges in distribution to the centred normal law with covariance matrix
where .
This is the multivariate central limit theorem for regenerative processes: several additive functionals of one regenerative process, observed over the same cycles, fluctuate jointly like a Gaussian vector, with covariances computed from a single cycle.
Formalization Note The random vectors are elements of the Euclidean space , and convergence is weak convergence of their laws to the Gaussian law (resp. ) along the real time . The matrices and are defined as covariance matrices, hence positive semidefinite; degenerate limits are allowed and no positivity is assumed. The hypothesis of the paper follows from .
import Mathlib import Definitions.Def_SmithRegenerative_CLT_CumulativeProcess open MeasureTheory ProbabilityTheory Filter Topology
namespace SmithRegenerative.CLT
/-- Smith (1955), Theorem 10, p. 30: "If `w_t^{(1)}, w_t^{(2)}, …, w_t^{(M)}` are M-cumulative
processes (based on the same sequence of regeneration points); if `μ₁ < ∞`, `E(Δ_n w̃_t^{(i)})² < ∞`,
and if `κ₁^{(i)} = EΔ_n w_t^{(i)}`, `a_kl = cov(Δ_n w_t^{(k)}, Δ_n w_t^{(l)}) < ∞`, for all `k, l` then
the random variables `(w_t^{(i)} − κ₁^{(i)} n_t)/(t/μ₁)^{1/2}` tend, as `t → ∞`, to be jointly
normally distributed with covariance matrix `a_kl`. If, in addition, `μ₂ < ∞`, then the random
variables `(w_t^{(i)} − (κ₁^{(i)}/μ₁)t)/(t/μ₁)^{1/2}` tend, as `t → ∞`, to be jointly normally
distributed with covariance matrix
`b_kl = cov{Δ_n(w_t^{(k)} − (κ₁^{(k)}/μ₁)t), Δ_n(w_t^{(l)} − (κ₁^{(l)}/μ₁)t)}`."
**Formalization Note** The `M` processes are `w i`, `i : Fin M`, on the common renewal process
`τ` (`t₀ = 0`, `w₀ = 0`), with i.i.d. cycle vectors (`IsCumulativeModel`). The random vectors
live in `EuclideanSpace ℝ (Fin M)`; "jointly normally distributed with covariance matrix `a`" is
convergence in distribution, along real `t → ∞`, to `multivariateGaussian 0 a`. `a` and `b` are
defined as covariance matrices (of `y₁^{(k)}`, resp. of `y₁^{(k)} − (κ₁^{(k)}/μ₁) t₁`, which is
`Δ_n(w^{(k)} − (κ₁^{(k)}/μ₁)t)`), so they are positive semidefinite and the limits are genuine,
possibly degenerate, Gaussian laws; no positivity is assumed. `E(Δ_n w̃^{(i)})² < ∞` is
integrability of `(ỹ₁^{(i)})²`; it makes every `a_kl` finite. `(t/μ₁)^{1/2}` is
`Real.sqrt (t / μ₁)`, and `n_t` is `count`. -/
theorem theorem_10 {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
{M : ℕ} (τ : ℕ → Ω → ℝ) (w : Fin M → ℝ → Ω → ℝ) (hw : IsCumulativeModel P τ w)
(hμ : Integrable (τ 1) P)
(hvar2 : ∀ i, Integrable (fun ω => varIncr (w i) τ 1 ω ^ 2) P) :
TendstoInDistribution
(fun (t : ℝ) (ω : Ω) => WithLp.toLp 2 (fun i : Fin M =>
(w i t ω - (∫ ω', incr (w i) τ 1 ω' ∂P) * (count τ t ω : ℝ)) /
Real.sqrt (t / mu1 P τ)))
atTop id (fun _ => P)
(multivariateGaussian 0 (covMatrix P (fun k => incr (w k) τ 1))) ∧
(Integrable (fun ω => τ 1 ω ^ 2) P →
TendstoInDistribution
(fun (t : ℝ) (ω : Ω) => WithLp.toLp 2 (fun i : Fin M =>
(w i t ω - (∫ ω', incr (w i) τ 1 ω' ∂P) / mu1 P τ * t) / Real.sqrt (t / mu1 P τ)))
atTop id (fun _ => P)
(multivariateGaussian 0 (covMatrix P (fun k ω =>
incr (w k) τ 1 ω - (∫ ω', incr (w k) τ 1 ω' ∂P) / mu1 P τ * τ 1 ω)))) := by sorry
end SmithRegenerative.CLT
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.