Wiener increments, quadratic variation and ∫W dW (Gardiner §4.2.2, §4.2.5) #
For a Wiener process W (Gardiner.Ito.IsWiener W P) and a partition
t₀ = p 0 ≤ p 1 ≤ … ≤ p n = t of [t₀, t] (Gardiner.Ito.IsPartition, with mesh ≤ δ),
write ΔWᵢ = W(p (i+1)) - W(p i) (Gardiner.Ito.dW) and Δtᵢ = p (i+1) - p i.
- Increments (Gardiner §3.8.1 / text before (4.2.16)): the
ΔWᵢare independent (Gardiner.Ito.IsWiener.dW_iIndepFun) andΔWᵢ ~ N(0, Δtᵢ)(Gardiner.Ito.IsWiener.dW_map). - Moments (4.2.14), (4.2.17), §4.2.5 i)–ii):
⟨ΔWᵢ²⟩ = Δtᵢ,⟨ΔWᵢ⁴⟩ = 3Δtᵢ²,⟨(ΔWᵢ² - Δtᵢ)²⟩ = 2Δtᵢ²(Gardiner.Ito.IsWiener.dW_sq_mean,…dW_four,…dW_sq_sub_sq). - Quadratic variation (4.2.14)–(4.2.19): with
Q = Σᵢ ΔWᵢ²(Gardiner.Ito.qvSum),⟨Q⟩ = t - t₀and the exact mean-square error⟨(Q - (t - t₀))²⟩ = 2 Σᵢ Δtᵢ² ≤ 2δ(t - t₀)(Gardiner.Ito.IsWiener.qv_mean,Gardiner.Ito.IsWiener.qv_ms_error,Gardiner.Ito.IsWiener.qv_ms_le), hencems-lim Q = t - t₀as the mesh tends to zero (Gardiner.Ito.IsWiener.qv_msLim). ∫ W dW(4.2.12)–(4.2.13), (4.2.20): the left-point sumS = Σᵢ W(p i) ΔWᵢ(Gardiner.Ito.itoSum) equals½[W(t)² - W(t₀)²] - ½Qfor every path (Gardiner.Ito.itoSum_eq, derived fromGardiner.SDE.ito_sum_identity), so⟨(S - ½[(W(t)² - W(t₀)²) - (t - t₀)])²⟩ = ½ Σᵢ Δtᵢ²(Gardiner.Ito.IsWiener.itoSum_ms_error) andms-lim S = ½[W(t)² - W(t₀)² - (t - t₀)](Gardiner.Ito.IsWiener.itoSum_msLim). This discharges, in mean square, the quadratic-variation hypothesisΣδᵢ² = t - t₀ofGardiner.SDE.ito_sum_of_qv(Gardiner.Ito.IsWiener.ito_sum_of_qv_ms).- Instances for the standard Wiener process (
Gardiner.Ito.stdWiener) on uniform partitions of[0, 1]:⟨(Q_n - 1)²⟩ = 2/nand⟨(S_n - ½(W(1)² - W(0)² - 1))²⟩ = 1/(2n)(Gardiner.Ito.qv_unif_instance,Gardiner.Ito.itoSum_unif_instance).
Erratum (4.2.18). The left-hand side is missing the square on Wᵢ - Wᵢ₋₁, and the middle
line Σ_{i,j}[(tᵢ - tᵢ₋₁) - (t - t₀)][(tⱼ - tⱼ₋₁) - (t - t₀)] should be
Σ_{i,j}(tᵢ - tᵢ₋₁)(tⱼ - tⱼ₋₁) - (t - t₀)², which vanishes. As printed, the middle line equals
(n - 1)²(t - t₀)², so the printed right-hand side 2Σ Δtᵢ² + (n - 1)²(t - t₀)² is wrong
(Gardiner.Ito.eq_4_2_18_printed_wrong: for n = 2 on [0, 1] the mean-square error is 1,
the printed expression gives 2). The final value 2 Σ Δtᵢ² is correct.
Partitions #
p 0 = t₀ ≤ p 1 ≤ … ≤ p n = t is a partition of [t₀, t] with mesh at most δ
(p is monotone on all of ℕ; its values beyond n are irrelevant).
- mono : Monotone p
Instances For
Σᵢ Δtᵢ² ≤ δ (t - t₀).
Increments #
The quadratic-variation sum Q = Σ_{i<n} ΔWᵢ².
Equations
- Gardiner.Ito.qvSum W p n ω = ∑ i ∈ Finset.range n, Gardiner.Ito.dW W p i ω ^ 2
Instances For
The left-point (Itô) sum S = Σ_{i<n} W(p i) ΔWᵢ of (4.2.12).
Equations
- Gardiner.Ito.itoSum W p n ω = ∑ i ∈ Finset.range n, W (p i) ω * Gardiner.Ito.dW W p i ω
Instances For
(4.2.12)–(4.2.13), for every path: S = ½[W(t)² - W(t₀)²] - ½ Σ ΔWᵢ². Derived from the
algebraic identity Gardiner.SDE.ito_sum_identity of the Itô-free layer.
ΔWᵢ ~ N(0, Δtᵢ).
The increments over a partition are independent.
§4.2.5 ii): ⟨(ΔWᵢ² - Δtᵢ)²⟩ = 2Δtᵢ².
The centred squared increments ΔWᵢ² - Δtᵢ.
(4.2.14): ⟨Σ ΔWᵢ²⟩ = t - t₀.
(4.2.15)–(4.2.18), corrected: the exact mean-square error of the quadratic-variation
sum, ⟨(Σ ΔWᵢ² - (t - t₀))²⟩ = 2 Σ Δtᵢ².
The mean-square error of the quadratic-variation sum is at most 2δ(t - t₀).
(4.2.19): ms-lim Σ ΔWᵢ² = t - t₀ along any sequence of partitions of [t₀, t] whose
mesh tends to zero.
The mean-square error of the left-point sum for ∫ W dW:
⟨(S - ½[(W(t)² - W(t₀)²) - (t - t₀)])²⟩ = ½ Σ Δtᵢ².
(4.2.20): ∫_{t₀}^t W dW = ms-lim Σ W(tᵢ₋₁)ΔWᵢ = ½[W(t)² - W(t₀)² - (t - t₀)] along any
sequence of partitions whose mesh tends to zero.
Comparison with Gardiner.SDE.ito_sum_of_qv. That theorem needs the exact identity
Σδᵢ² = t - t₀, which never holds for Wiener increments (it is a statement about a random
variable). With real Wiener increments, ito_sum_of_qv holds up to an error whose mean square is
exactly ½ Σ Δtᵢ²: for the increments δᵢ = ΔWᵢ(ω),
Σ psum δ i · δ i = ½[(psum δ n)² - (t - t₀)] + Eₙ(ω) with ⟨Eₙ²⟩ = ½ Σ Δtᵢ² ≤ ½δ(t - t₀).
Instances: the standard Wiener process on uniform partitions of [0, 1] #
Instance of (4.2.18)–(4.2.19): for the standard Wiener process and the uniform partition of
[0, 1] into n cells, ⟨(Σ ΔWᵢ² - 1)²⟩ = 2/n, which tends to 0.
Instance of (4.2.20): for the standard Wiener process and the uniform partition of [0, 1],
⟨(Σ W(tᵢ₋₁)ΔWᵢ - ½[W(1)² - W(0)² - 1])²⟩ = 1/(2n).
The uniform partitions realise ms-lim Σ ΔWᵢ² = t - t₀ for the standard Wiener process.
Erratum (4.2.18) #
The right-hand side of (4.2.18) as printed:
2 Σᵢ Δtᵢ² + Σ_{i,j} [Δtᵢ - (t - t₀)][Δtⱼ - (t - t₀)].
Equations
- Gardiner.Ito.printed_4_2_18 p n = 2 * ∑ i ∈ Finset.range n, (p (i + 1) - p i) ^ 2 + ∑ i ∈ Finset.range n, ∑ j ∈ Finset.range n, (p (i + 1) - p i - (p n - p 0)) * (p (j + 1) - p j - (p n - p 0))
Instances For
Erratum (4.2.18). For the uniform partition of [0, 1] into two cells the mean-square
error of the quadratic-variation sum is 1, but the printed right-hand side of (4.2.18) gives 2.
The correct value is 2 Σ Δtᵢ² (Gardiner.Ito.IsWiener.qv_ms_error).