The Wiener integral of deterministic functions (Gardiner §4.2.1, §4.2.6) #
For f : ℝ → ℝ and 0 ≤ a ≤ b, Gardiner.Ito.wienerInt W P f a b = ∫ₐᵇ f(s) dW(s) is the
mean-square limit of the left-point Riemann sums Σᵢ f(tᵢ)(W(tᵢ₊₁) - W(tᵢ)) on the uniform
partitions of [a, b] (Gardiner.Ito.riemannSum, (4.2.8) with τᵢ = tᵢ₋₁). Each Riemann sum is
the simple integral of a deterministic simple integrand (Gardiner.Ito.riemannSum_eq). For
continuous f:
- the limit exists (
Gardiner.Ito.IsWiener.wienerInt_spec); - mean zero (4.2.40) and covariance
⟨∫ₐᵇ f dW ∫_c^d g dW⟩ = ∫ f g 1_{(a,b]∩(c,d]}, in particular∫ₐ^{min(b,d)} f gfor a common left endpoint and the isometry⟨(∫ₐᵇ f dW)²⟩ = ∫ₐᵇ f²(Gardiner.Ito.IsWiener.wienerInt_mean,…wienerInt_mul,…wienerInt_mul_same,…wienerInt_sq); - Gaussian:
∫ₐᵇ f dW ~ N(0, ∫ₐᵇ f²)(Gardiner.Ito.IsWiener.wienerInt_map), and pairs on a common interval are jointly Gaussian (Gardiner.Ito.IsWiener.wienerInt_hasGaussianLaw_pair, fromGardiner.Ito.hasGaussianLaw_pair); - linear (
Gardiner.Ito.IsWiener.wienerInt_add_smul,…wienerInt_lin_ae), and∫ₐᵇ dW = W(b) - W(a)(Gardiner.Ito.IsWiener.wienerInt_one); - integration by parts for
C¹f:∫ₐᵇ f dW = f(b)W(b) - f(a)W(a) - ∫ₐᵇ f'(s)W(s) dsalmost surely, linking the integral to the pathwise layer (Gardiner.Ito.IsWiener.wienerInt_ibp).
For step functions the same definition gives the simple integral
(SHSLib.Gardiner.Ito.WienerStep). Instances for stdWiener: SHSLib.Gardiner.Ito.WienerInstances.
Cells of a partition #
Uniform partitions #
The uniform partition of [a, b] into n + 1 cells.
Equations
- Gardiner.Ito.grid a b n = Gardiner.Ito.unifPart a b (n + 1)
Instances For
Uniform continuity on the cells: for g continuous on [a, b] and n large, g
oscillates by at most ε on every cell of the uniform partition.
Step approximations and Riemann sums #
The left-point step approximation of f on the uniform partition of [a, b],
fₙ(s) = Σᵢ f(tᵢ) 1_{(tᵢ, tᵢ₊₁]}(s).
Equations
- Gardiner.Ito.stepFn f a b n s = ∑ i ∈ Finset.range (n + 1), f (Gardiner.Ito.grid a b n i) * Gardiner.Ito.ind (Gardiner.Ito.grid a b n i) (Gardiner.Ito.grid a b n (i + 1)) s
Instances For
The left-point Riemann sum Σᵢ f(tᵢ)(W(tᵢ₊₁) - W(tᵢ)) on the uniform partition of [a, b]
into n + 1 cells (Gardiner (4.2.8) with τᵢ = tᵢ₋₁).
Equations
- Gardiner.Ito.riemannSum W f a b n ω = ∑ i ∈ Finset.range (n + 1), f (Gardiner.Ito.grid a b n i) * (W (Gardiner.Ito.grid a b n (i + 1)) ω - W (Gardiner.Ito.grid a b n i) ω)
Instances For
The step approximation converges uniformly: for continuous f and large n,
|fₙ(s) - f(s) 1_{(a,b]}(s)| ≤ ε 1_{(a,b]}(s) for every s.
f 1_{(a,b]} is square integrable for continuous f.
fₙ → f 1_{(a,b]} in L²(ds).
Riemann sums as simple integrals #
The deterministic simple integrand Σᵢ f(tᵢ) 1_{(tᵢ, tᵢ₊₁]} on the uniform partition.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The isometry for Riemann sums of deterministic functions:
⟨Rₙ(f; a, b) Rₘ(g; c, d)⟩ = ∫ fₙ gₘ ds.
The Riemann sums of a continuous function form a mean-square Cauchy sequence.
A linear combination of Wiener increments over a partition is centred Gaussian.
The Wiener integral #
The Wiener integral ∫ₐᵇ f(s) dW(s) of a deterministic function: the mean-square limit of
the left-point Riemann sums on the uniform partitions of [a, b] (Gardiner (4.2.8), (4.2.10)).
For continuous f and 0 ≤ a ≤ b the limit exists (Gardiner.Ito.IsWiener.wienerInt_spec).
Equations
- Gardiner.Ito.wienerInt W P f a b = Gardiner.Ito.msLim (Gardiner.Ito.riemannSum W f a b) P
Instances For
The Wiener integral of a continuous function exists: it is square integrable and the Riemann sums converge to it in mean square.
The mean of the Wiener integral vanishes (4.2.40).
The covariance of two Wiener integrals (Gardiner §4.4.1, via (4.2.42)):
⟨∫ₐᵇ f dW ∫_c^d g dW⟩ = ∫ f 1_{(a,b]} g 1_{(c,d]} ds.
The covariance for a common left endpoint:
⟨∫ₐᵇ f dW ∫ₐᵈ g dW⟩ = ∫ₐ^{min(b,d)} f g ds.
The Itô isometry for the Wiener integral: ⟨(∫ₐᵇ f dW)²⟩ = ∫ₐᵇ f² ds.
The Wiener integral is Gaussian: ∫ₐᵇ f dW ~ N(0, ∫ₐᵇ f² ds).
The mean square of a linear combination of Wiener integrals is the L²(ds) norm of the
corresponding combination of integrands.
Linear relations transfer from integrands to Wiener integrals: if
Σᵢ cᵢ Fᵢ 1_{(Aᵢ, Bᵢ]} = 0 pointwise, then Σᵢ cᵢ ∫_{Aᵢ}^{Bᵢ} Fᵢ dW = 0 almost surely.
Linearity of the Wiener integral: ∫ (αf + βg) dW = α ∫ f dW + β ∫ g dW a.s.
The Wiener integral of the constant 1 is the increment: ∫ₐᵇ dW = W(b) - W(a).
A pair of random variables is jointly Gaussian as soon as every linear combination is Gaussian.
Joint law: Wiener integrals over a common interval are jointly Gaussian.
Integration by parts #
Pathwise integration by parts for Riemann sums: for a C¹ function f and a continuous
path w, Σᵢ f(tᵢ)(w(tᵢ₊₁) - w(tᵢ)) → f(b)w(b) - f(a)w(a) - ∫ₐᵇ f'(s) w(s) ds.
Integration by parts for the Wiener integral of a C¹ function:
∫ₐᵇ f dW = f(b)W(b) - f(a)W(a) - ∫ₐᵇ f'(s) W(s) ds almost surely. This identifies the
mean-square integral with an expression in the paths of W.