Proposition 4.1, A1 — the two forms of assumption A1 are equivalent
ProvedStochApproxDyn.Interpolation.a1At_iff_tendsto_deltap2o-batch-pfp1ap2o-gran-per-chapterp2o-plan-paperp2o-v1stochastic-approximation
Let the step sizes satisfy , , , let be a sequence in , and let , , and be as in §4.1. Then for every ,
The paper states assumption A1 in the discrete form and asserts that the continuous form is equivalent; Propositions 4.2 and 4.4 verify the discrete form, while the proof of Proposition 4.1 uses the continuous one.
Formalization Note The discrete form is written with : for all there is such that the norm of every partial sum with and is at most . The equivalence is stated for each fixed .
Preamble
import Mathlib import Definitions.Def_StochApproxDyn_Interpolation_Scheme open scoped NNReal Topology open Filter
Formal statement
namespace StochApproxDyn.Interpolation
/-- Benaïm 1999, Proposition 4.1, assumption A1, p. 12 ("or equivalently"): under the standing
assumptions on `γ`, for every `T > 0` the sum form of A1,
`lim_{n→∞} sup{‖Σ_{i=n}^{k−1} γ_{i+1} U_{i+1}‖ : k = n + 1, …, m(τ_n + T)} = 0`,
holds if and only if `lim_{t→∞} Δ(t, T) = 0`, where `Δ(t, T) = sup_{0≤h≤T} ‖∫_t^{t+h} Ū(s) ds‖`. -/
theorem a1At_iff_tendsto_delta {d : ℕ} (γ : ℕ → ℝ) (hγ : IsStepSizeSeq γ)
(U : ℕ → EuclideanSpace ℝ (Fin d)) (T : ℝ) (hT : 0 < T) :
A1At γ U T ↔ Tendsto (fun t : ℝ => Delta γ U t T) atTop (𝓝 0) := by sorry
end StochApproxDyn.Interpolation
Source
Benaïm, Dynamics of Stochastic Approximation Algorithms, Séminaire de Probabilités XXXIII, LNM 1709 (1999), p. 12, Proposition 4.1, assumption A1 ("or equivalently") and Eq. (10)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.