Setting for the differential Chapman–Kolmogorov equation (Gardiner §3.4) #
Formalization of the derivation of the differential Chapman–Kolmogorov equation following C. W. Gardiner, Handbook of Stochastic Methods, §3.4 and §3.4.1.
This file sets up the state space, partial derivatives, the second order Taylor remainder (3.4.10)–(3.4.11), the three regularity conditions (i), (ii), (iii) of §3.4 [(3.4.1), (3.4.2), (3.4.3)], and the four quantities into which the difference quotient (3.4.8)–(3.4.9) is split in (3.4.12).
References #
- [Gar04], §3.4
The state space ℝⁿ of the process, with its Euclidean norm.
Equations
- Gardiner.Sp n = EuclideanSpace ℝ (Fin n)
Instances For
Partial derivatives #
pd f i z is the partial derivative ∂f(z)/∂z_i.
Equations
- Gardiner.pd f i z = (fderiv ℝ f z) (EuclideanSpace.single i 1)
Instances For
pd2 f i j z is the second partial derivative ∂²f(z)/∂z_i ∂z_j.
Equations
- Gardiner.pd2 f i j z = Gardiner.pd (Gardiner.pd f j) i z
Instances For
The Taylor remainder (3.4.10), (3.4.11) #
The first two nonconstant terms of the Taylor expansion (3.4.10).
Equations
Instances For
The two regions of integration #
The set of pairs (x, z) with |x - z| < ε.
Equations
- Gardiner.nearDiag ε = {q : Gardiner.Sp n × Gardiner.Sp n | dist q.1 q.2 < ε}
Instances For
The set of pairs (x, z) with |x - z| ≥ ε.
Equations
- Gardiner.farDiag ε = {q : Gardiner.Sp n × Gardiner.Sp n | ε ≤ dist q.1 q.2}
Instances For
Conditions (i), (ii), (iii) of §3.4 #
Condition (i), (3.4.1): p(x, t+Δt | z, t)/Δt → W(x|z,t) uniformly in x, z
on |x - z| ≥ ε, for every ε > 0. Here k dt z x stands for
p(x, t + dt | z, t) and W x z for W(x | z, t).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Condition (ii), (3.4.2): the first moment over the ε-ball produces the drift
vector A_i(z, t) up to an error O(ε) which is uniform in z.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Condition (iii), (3.4.3): the second moment over the ε-ball produces the
diffusion matrix B_ij(z, t) up to an error O(ε) which is uniform in z.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The quantities appearing in the derivation #
Throughout, p s z denotes p(z, s | y, t') for a fixed initial condition
(y, t'), and k dt z x denotes the transition density p(x, t + dt | z, t).
The weight p(x, t + Δt | z, t) p(z, t | y, t') occurring in every double
integral of the derivation, as a function of the pair q = (x, z).
Equations
- Gardiner.kernelWeight p k t dt q = k dt q.2 q.1 * p t q.2
Instances For
(3.4.8): the difference quotient whose limit as Δt → 0 is
∂_t ∫ dx f(x) p(x, t | y, t').
Equations
- Gardiner.diffQuot p f t dt = ((∫ (x : Gardiner.Sp n), f x * p (t + dt) x) - ∫ (z : Gardiner.Sp n), f z * p t z) / dt
Instances For
Lines 1 and 2 of (3.4.12): the contribution of the first and second order
Taylor terms on the region |x - z| < ε.
Equations
- Gardiner.Q12 p k f t dt ε = 1 / dt * ∫ (q : Gardiner.Sp n × Gardiner.Sp n) in Gardiner.nearDiag ε, Gardiner.taylorPoly f q.2 q.1 * Gardiner.kernelWeight p k t dt q
Instances For
Line 3 of (3.4.12): the remainder term.
Equations
- Gardiner.Q3 p k f t dt ε = 1 / dt * ∫ (q : Gardiner.Sp n × Gardiner.Sp n) in Gardiner.nearDiag ε, Gardiner.taylorRem f q.2 q.1 * Gardiner.kernelWeight p k t dt q
Instances For
Lines 4–6 of (3.4.12), combined as explained in the text: the contribution of
the region |x - z| ≥ ε.
Equations
- Gardiner.Qjump p k f t dt ε = 1 / dt * ∫ (q : Gardiner.Sp n × Gardiner.Sp n) in Gardiner.farDiag ε, (f q.1 - f q.2) * Gardiner.kernelWeight p k t dt q
Instances For
The second moment appearing in the estimate (3.4.14) of line 3.
Equations
- Gardiner.Qsq p k t dt ε = 1 / dt * ∫ (q : Gardiner.Sp n × Gardiner.Sp n) in Gardiner.nearDiag ε, ‖q.1 - q.2‖ ^ 2 * Gardiner.kernelWeight p k t dt q
Instances For
(3.4.13): the drift and diffusion part of the right-hand side.
Equations
- Gardiner.driftDiffTerm p A B f t = ∫ (z : Gardiner.Sp n), (∑ i : Fin n, A z i * Gardiner.pd f i z + 1 / 2 * ∑ i : Fin n, ∑ j : Fin n, B z i j * Gardiner.pd2 f i j z) * p t z
Instances For
The Δt → 0 limit of Qjump, before the change of variables x ↔ z.
Equations
- Gardiner.jumpPre p W f t ε = ∫ (q : Gardiner.Sp n × Gardiner.Sp n) in Gardiner.farDiag ε, (f q.1 - f q.2) * (W q.1 q.2 * p t q.2)
Instances For
(3.4.15): the jump contribution, in the form in which it enters (3.4.16).
Equations
- Gardiner.jumpTerm p W f t ε = ∫ (z : Gardiner.Sp n), f z * ∫ (x : Gardiner.Sp n) in (Metric.ball z ε)ᶜ, W z x * p t x - W x z * p t z