Eq. (17):
ProvedStochApproxDyn.MartingaleNoise.summable_eq17martingalesp2o-batch-pfp1ap2o-gran-per-chapterp2o-plan-paperp2o-v1stochastic-approximation
Let , given by (7), be a Robbins–Monro algorithm with noise in and step sequence . Suppose that for some
Then for every
where is the noise deviation (10) of the realised noise path.
By the Borel–Cantelli lemma this gives almost surely as .
Formalization Note The paper prints the chain , whose middle equality drops the factor ; only the finiteness is formalized. Expectations are integrals in and the sum is the sum in ; the summability of is real summability of a nonnegative sequence.
Preamble
import Mathlib import Definitions.Def_StochApproxDyn_MartingaleNoise_Interpolation import Definitions.Def_StochApproxDyn_MartingaleNoise_AssumptionA1 import Definitions.Def_StochApproxDyn_MartingaleNoise_RobbinsMonro open MeasureTheory ProbabilityTheory Filter Topology open scoped ENNReal
Formal statement
namespace StochApproxDyn.MartingaleNoise
/-- Benaïm 1999, §4.2, proof of Proposition 4.2, Eq. (17), p. 16: under the hypotheses of
Proposition 4.2, for every `T > 0`, `∑_{k ≥ 0} E(Δ(kT, T)^q) < ∞`. -/
theorem summable_eq17 {d : ℕ} {Ω : Type*} {m0 : MeasurableSpace Ω}
(P : Measure Ω) [IsProbabilityMeasure P] (ℱ : Filtration ℕ m0)
(F : EuclideanSpace ℝ (Fin d) → EuclideanSpace ℝ (Fin d)) (γ : ℕ → ℝ)
(x U : ℕ → Ω → EuclideanSpace ℝ (Fin d)) (hRM : IsRobbinsMonro P ℱ F γ x U)
(q : ℝ) (hq : 2 ≤ q)
(hmom : ∃ M : ℝ≥0∞, M < ⊤ ∧ ∀ n : ℕ, ∫⁻ ω, ‖U (n + 1) ω‖ₑ ^ q ∂P ≤ M)
(hsum : Summable (fun n : ℕ => γ (n + 1) ^ (1 + q / 2))) :
∀ T : ℝ, 0 < T →
∑' k : ℕ, ∫⁻ ω, noiseDev γ (fun n => U n ω) ((k : ℝ) * T) T ^ q ∂P < ⊤ := by sorry
end StochApproxDyn.MartingaleNoise
Source
Benaïm, Dynamics of Stochastic Approximation Algorithms, Séminaire de Probabilités XXXIII, LNM 1709 (1999), DOI 10.1007/BFb0096509, Section 4.2, proof of Proposition 4.2, p. 16 (PDF p. 17), Eq. (17)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.