5
$\begingroup$

Let

  • $b,\sigma:\mathbb R\to\mathbb R$ be differentiable and Lipschitz continuous
  • $(\Omega,\mathcal A,\operatorname P)$ be a probability space
  • $(\mathcal F_t)_{t\ge0}$ be a complete and right-continuous filtration on $(\Omega,\mathcal A,\operatorname P)$
  • $(W_t)_{t\ge0}$ be an $\mathcal F$-Brownian motion on $(\Omega,\mathcal A,\operatorname P)$
  • $(X^x_t)_{(t,\:x)\in[0,\:\infty)\times\mathbb R}$ be a real-valued continuous $\mathcal F$-adapted process on $(\Omega,\mathcal A,\operatorname P)$ with $$X^x_t=x+\int_0^tb(X^x_s)\:{\rm d}s+\int_0^t\sigma(X^x_s)\:{\rm d}W_s\tag1$$ for all $t\ge0$ almost surely for all $x\in E$

Are we able to show that $\mathbb R\ni x\mapsto X^x_t(\omega)$ is differentiable for all $(\omega,t)\in(\Omega\setminus N)\times[0,\infty)$ for some $\operatorname P$-null set $N$?

Note that the hypotheses are sufficient to guarantee the existence of the process $X$ above. However, I'm not sure if they are sufficient to guarantee the existence of a real-valued continuous $\mathcal F$-adapted process $(Y^x_t)_{(t,\:x)\in[0,\:\infty)\times\mathbb R}$ with $$Y^x_t=\int_0^tb'(X^x_s)Y^x_s\:{\rm d}s+\int_0^t\sigma'(X^x_s)Y^x_s\:{\rm d}W_s\tag2$$ for all $t\ge0$ almost surely for all $x\in\mathbb R$ as well. However, if this process $Y$ exists, then it's clearly a candidate for the desired process obtained by differentiating $X$ with respect to the spatial parameter.

Maybe it's useful to remember how the existence of $X$ is proved: For fixed $T>0$ and $\xi\in L^2(\operatorname P)$, we know that the map $$\Xi_T(Z):=\xi+\left(\int_0^tb(s,Z_s)\:{\rm d}s\right)_{t\in[0,\:T]}+\left(\int_0^t\sigma(s,Z_s)\:{\rm d}W_s\right)_{t\in[0,\:T]}\in\mathcal B_T$$ for $Z\in\mathcal B_T$, where $$\mathcal C_T:=\left\{Z:Z\text{ is a continuous }(\mathcal F_t)_{t\in[0,\:T]}\text{-adapted process on }(\Omega,\mathcal A,\operatorname P)\right\}$$ is equipped with $$\left\|Z\right\|_T:=\left\|\sup_{t\in[0,\:T]}Z_t\right\|_{L^2(\operatorname P)}\;\;\;\text{for }Z\in\mathcal C_T$$ and $$\mathcal B_T:=\left\{Z\in\mathcal C_T:\left\|Z\right\|_T<\infty\right\},$$ has a unique fixed point and there is a constant $c\ge0$, depending only on $b$ and $\sigma$, with $$\left\|\Xi_T(Z)\right\|_T^2\le3\left(\operatorname E\left[|\xi|^2\right]+c(T+4)\left(T+\int_0^T\left\|Z\right\|_t^2\:{\rm d}t\right)\right)\tag3$$ for all $Z\in\mathcal B_T$.

Maybe this can be used to show that $$\left\|\frac{X^{x+h}-X^x}h-Y^x\right\|_T\le c_1|h|\;\;\;\text{for all }T>0\text{ and }h\in\mathbb R\tag4$$ for some $c_1\ge0$, depending only on $b$ and $\sigma$, which should even allow for a stronger conclusion then the one asked for in the question.

$\endgroup$
3
  • 1
    $\begingroup$ Have you looked at Kunita's st flour lecture notes on flows of diffeomorphism? $\endgroup$ Jun 9, 2020 at 15:10
  • $\begingroup$ @oferzeitouni Yes, sure. But Kunita is hard to read and I think the presented theory is overkill for this rather simple scenario. $\endgroup$
    – 0xbadf00d
    Jun 9, 2020 at 15:11
  • $\begingroup$ I looked at Kunita's lecture and the subject does not seem addressed. Indeed Kunita is hard to read. To further your question, I was wondering if a kind of Taylor's expansion can be used like : $X_t^{x_1}(\omega) = X_t^{0}(\omega) + (x_1 - 0) \frac{\partial X_t^x}{\partial x}(x_1)$ $\endgroup$ Mar 15 at 11:29

1 Answer 1

1
$\begingroup$

As mentioned in the comments, Kunita's lectures cover this eg. Lectures on Stochastic Flows And Applications, i.e. depending on the regularity of the coefficients we have analogous differentiability in initial conditions

enter image description here

Another source is in Varadhan's notes too Stochastic1

enter image description here

As you mentioned, indeed the proof goes through creating new systems satisfied by the derivative processes and then showing existence-uniqueness for the entire system.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.