Section 4.1, Eq. (9) — the interpolated process solves the integral form of (7)
ProvedStochApproxDyn.Interpolation.interpAffine_sub_eq_integralp2o-batch-pfp1ap2o-gran-per-chapterp2o-plan-paperp2o-v1stochastic-approximation
Let be continuous, let the step sizes satisfy , , , and let follow the scheme . Let be the affine interpolation of on the time scale , the piecewise constant interpolation and the piecewise constant interpolation of the perturbations. Then for every
Equation (9) turns the discrete recursion into an integral equation that differs from the integral form of only by the replacement of with and by the noise integral; every later comparison with the flow of starts from it.
Formalization Note The integral is the interval integral of a piecewise constant function, which is integrable on under the standing assumptions, so no junk value enters.
Preamble
import Mathlib import Definitions.Def_StochApproxDyn_Interpolation_Scheme open scoped NNReal Topology open Filter
Formal statement
namespace StochApproxDyn.Interpolation
/-- Benaïm 1999, §4.1, Eq. (9), p. 12: under the standing assumptions on `γ` and the recursion (7),
the affine interpolated process solves the integral form of (7):
`X(t) − X(0) = ∫_0^t [F(X̄(s)) + Ū(s)] ds` for every `t ≥ 0`. -/
theorem interpAffine_sub_eq_integral {d : ℕ}
(F : EuclideanSpace ℝ (Fin d) → EuclideanSpace ℝ (Fin d)) (hF : Continuous F)
(γ : ℕ → ℝ) (hγ : IsStepSizeSeq γ) (x U : ℕ → EuclideanSpace ℝ (Fin d))
(hx : SatisfiesScheme F γ x U) (t : ℝ) (ht : 0 ≤ t) :
interpAffine γ x t - interpAffine γ x 0 =
∫ s in (0 : ℝ)..t, (F (interpConst γ x s) + noiseInterp γ U s) := by sorry
end StochApproxDyn.Interpolation
Source
Benaïm, Dynamics of Stochastic Approximation Algorithms, Séminaire de Probabilités XXXIII, LNM 1709 (1999), p. 12, Section 4.1, Eq. (9)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.