Existence of solutions of Itô SDEs: Picard iteration #
Gardiner §4.3.1, (4.3.10)–(4.3.11). Under the Lipschitz and growth conditions on [t₀, T]
(SDECoeffs a b t₀ T K), for every horizon T and every square integrable, 𝓕 t₀-measurable
initial value ξ, the Picard iterates
x₀ = ξ, x_{n+1}(t) = ξ + ∫_{t₀}^t a(x_n(s), s) ds + ∫_{t₀}^t b(x_n(s), s) dW(s)
(taken in continuous adapted versions) converge almost surely uniformly on [t₀, T], and in
E sup |·|², to a solution of (4.3.4).
picardStep,picardSeq,sdeSol: the iteration and its limit;IsRegProc: jointly measurable, adapted, a.s. continuous on[t₀, T], withE sup_{[t₀, T]} |x|² < ∞;isSDESol_sdeSol: existence:sdeSolis a solution of (4.3.4) on[t₀, T].
A regular process on [t₀, T]: jointly measurable, adapted on [t₀, T], almost surely
continuous on [t₀, T], with E sup_{t₀ ≤ t ≤ T} |x(t)|² < ∞.
- measurable : Measurable (Function.uncurry x)
- adapted (t : ℝ) : t ∈ Set.Icc t₀ T → MeasureTheory.StronglyMeasurable (x t)
Instances For
Processes with E sup |x - y|² = 0 are indistinguishable.
E sup |x - z|² ≤ 2 E sup |x - y|² + 2 E sup |y - z|².
Elementary series bounds #
The Picard step #
The Picard map: x ↦ ξ + ∫_{t₀}^· a(x(s), s) ds + ∫_{t₀}^· b(x(s), s) dW(s), taken in its
adapted, continuous version.
Equations
- Gardiner.Ito.StochDE.picardStep W 𝓕 P a b t₀ T ξ x t ω = ξ ω + Gardiner.Ito.StochDE.driftVer a x t₀ t ω + Gardiner.Ito.StochDE.itoCont W 𝓕 P T (Gardiner.Ito.StochDE.drive t₀ b x) t ω
Instances For
The Picard iterates x₀ = ξ, x_{n+1} = picardStep x_n.
Equations
- Gardiner.Ito.StochDE.picardSeq W 𝓕 P a b t₀ T ξ 0 = fun (x : ℝ) (ω : Ω) => ξ ω
- Gardiner.Ito.StochDE.picardSeq W 𝓕 P a b t₀ T ξ n.succ = Gardiner.Ito.StochDE.picardStep W 𝓕 P a b t₀ T ξ (Gardiner.Ito.StochDE.picardSeq W 𝓕 P a b t₀ T ξ n)
Instances For
The solution of the SDE (4.3.4): the pointwise limit of the Picard iterates.
Equations
- Gardiner.Ito.StochDE.sdeSol W 𝓕 P a b t₀ T ξ t ω = limUnder Filter.atTop fun (n : ℕ) => Gardiner.Ito.StochDE.picardSeq W 𝓕 P a b t₀ T ξ n t ω
Instances For
The Picard step of a regular process is regular and satisfies the integral equation.
The Picard step is Lipschitz for the E sup |·|² distance (with Gronwall rate).
The initial iterate x₀ = ξ is regular.
All Picard iterates are regular.
Convergence of the Picard iterates #
sup_{t₀ ≤ u ≤ T} |x_{n+1}(u) - x_n(u)|.
Equations
- One or more equations did not get rendered due to their size.
Instances For
E sup |x₁ - x₀|².
Equations
- One or more equations did not get rendered due to their size.
Instances For
The Picard bound E sup_{[t₀, t]} |x_{n+1} - x_n|² ≤ M (L (t - t₀))ⁿ / n!.
E (∑_{k ≥ n} sup |x_{k+1} - x_k|)² ≤ 2 M e^{2 L (T - t₀)} (L (T - t₀))ⁿ / n!.
Almost surely, the Picard iterates converge uniformly on [t₀, T] to sdeSol, which is
continuous there, with the tail bound |x(t) - x_n(t)| ≤ ∑_{k ≥ n} sup |x_{k+1} - x_k|.
Mean-square convergence of the Picard iterates:
E sup_{[t₀, T]} |x - x_n|² ≤ 2 M e^{2 L (T - t₀)} (L (T - t₀))ⁿ / n!.
The limit of the Picard iterates is a regular process.
The limit is a fixed point of the Picard map, up to indistinguishability.
Existence (Gardiner §4.3.1): under the Lipschitz and growth conditions (4.3.10)–(4.3.11)
on [t₀, T], for every horizon T ≥ t₀ ≥ 0 and every square integrable, 𝓕 t₀-measurable
initial value ξ, the limit sdeSol of the Picard iterates is a continuous adapted solution of
(4.3.4) on [t₀, T].
Existence and uniqueness (Gardiner §4.3.1): there is a continuous adapted solution of
(4.3.4) on [t₀, T], and any two solutions are indistinguishable.