Article Title: Stochastic Motion Under G-Framework:
I. Nelson Stochastic Derivatives
Authors(s): Hong Zhang and D. Kannan
Journal: Stochastic Analysis and Applications
Bibliometrics: Published in Volume 32, Issue 6, pp. 1067–1096.
DOl: 10.1080/07362994.2014.964870
Publisher: Taylor & Francis
In our earlier paper, we were working on stochastic motion under the sublinear expectation framework. However, in the proof of Theorem 3.6 there, we made a careless mistake in working with , when
. Subsequently, we need to redefine the function ξθ and this affects the definitions of
and
Consequently, expressions/formulas in two other results and two examples change, and the corrected formulas are given below.
Assumption K-2:
For each , let pθ be the probability density function of Xt that satisfies Equation (3.5) under Pθ. Let
Define
Assume that the following functions are continuous and in L2G(Ω):
Note: Theorem 3.6, Theorem 3.7, Proposition 3.11, Example 4.4, and Example 4.5 (given below) all discuss the one-dimensional G-Itô process Xt satisfying
under the Assumptions (K-1) and (K-2). Here, X0 ∈ L2G(Ω0), and α, β, η ∈ M2G(0, T) are bounded processes. We will not include this hypothesis in the statements of the said theorems and examples below.
Theorem 3.6
Let Xt be the one dimensional G −Itô process
where X0 ∈ L2G(Ω0), α, β, η ∈ M2G(0, T) are bounded processes. Assume the conditions (K-1) and (K-2). Then
(1)
(2)
(3)
(4)
Proof.
For t ∈ (0, T), and h > 0, satisfying 0 < t − h < t < t + h < T, we have
Observing that the process ∫t + htη(u) d⟨B⟩u − 2∫t + htG(η(u)) du is a G-martingale with respect to the filtration
, and that
we have
Hence,
Similarly,
Recall now that a one-dimensional G-Itô process Xt admits the following representation under the probability measure Pθ,
and that the time-reversed process
of X(t) under Pθ satisfies
here,
is a
-Brownian motion under Pθ in which
and, with pθ(t) denoting the probability density of Xt under Pθ,
For fixed t ∈ [0, T] and
, set
It is an
-martingale for each θ. For 0 ≤ s < t ≤ T,
and
For each fixed ,
and
is an
-martingale under Pθ with
and
Notice that for h > 0, [t − h, t]⊂[0, T], and fixed , we have
Letting h → 0+, we now get
and, hence,
Consequently,
Similarly, we obtain
Noticing now that,
we obtain
(5)
and
It is easily seen that DiX(t), i = 1, 2, 3, 4, are continuous on (0, T) in L2G(Ω).
Recall the hypothesis in the Note above. Under that hypothesis, we have the following Proposition as a special case of the above theorem.
Proposition 3.7
now reads as: If α(t) = α, η(t) = η, β(t) = β are bounded in L2G(Ω0), η is deterministic, and X0 ∈ L2G(Ω0), then DiX, i = 1,…, 4, exist, are continuous on (0, T), and are given by:
Theorem 3.11
now reads as: Subject to the Note above, if with
we have
Proof.
Note from Itô’s formula that, for s, t ∈ [0, T] with s < t,
and
The results follow from Theorem 3.6.
Now the and
of Example 4.2, are given by:
Example 4.4:
Keeping in mind the Note supra, the forward and backward drifts are given by
respectively, and the current and osmotic velocities of X(t) are given by
respectively.
Example 4.5:
Now, for with
, we have the forward and backward drifts of ϕ(X(t)) given by
respectively, and the current and osmotic velocities of ϕ(X(t)) are given by
respectively.
Next we make the corresponding changes in the multidimensional case. Let B be a d-dimensional G-Brownian motion and X = (X1,…, Xn)T be an n-dimensional process on [0, T] in the form
(6) Here, for ν = 1, 2,…, n, i, j = 1, 2,…, d, αν, ηνij, βνj ∈ M2G(0, T) are bounded and continuous processes, X(0) is a given n-dimensional random column vector in L2G(Ω),β = (β1, β2,…, βn) is an d × n-matrix with d-dimensional column vectors βν,ν = 1, 2,…, n, and β* is the transpose of β.
Assumption K-3:
For each , let pθ be the probability density function of Xt under Pθ. For the d × d matrices
we set
Assume that
are continuous in L2(Ω) .
Theorem 5.1:
Let B be an d-dimensional G-Brownian motion and X = (X1,…, Xn)T be an n-dimensional process on [0, T] in the form
(7)
Here, for ν = 1, 2,…, n, i, j = 1, 2,…, d, αν, ηνij, βνj ∈ M2G(0, T) are bounded and continuous processes, and X(0) = (X1(0),…, Xn(0))T is a given random vector in L2G(Ω). β = (β1, β2,…, βn) is an d × n-matrix with d-dimensional column vectors βν,ν = 1, 2,…, n,, and β* is the transpose of β. We then have the following:
1. | DiX(t), i = 1, 2, exist, are continuous on (0, T), and are given by
| ||||
2. | If in addition, the Assumptions (K-1) and (K-3) also hold, then DiX(t), i = 3, 4, exist, are continuous on (0, T), and are given by
|
Proof.
(2) Notice that the time reversed process of X(t) under Pθ satisfies
Here,
is a d-dimensional
-Brownian motion under Pθ,
and
where pθ(t) is the probability density of Xt under Pθ.
For h > 0, t, t − h ∈ [0, T], we have and
For each
,
is an
-martingale, and
We have
Therefore,
and, consequently,
Similarly,
Theorem 5.2
now reads: Assume that with
for μ, ν = 1,…, n. Then,
Proof.
This follows from Itô’s formula and Theorem 5.1.