-1
$\begingroup$

Let $(\Omega,\mathcal A,\operatorname P)$ be a probability space, $(B_t)_{t\ge 0}$ be a Brownian motion on $(\Omega,\mathcal A,\operatorname P)$ and $f\in C^1(\mathbb R_{\ge 0})$. Can we show that $$\int_0^tf(s){\rm d}B_s=-\int_0^tf'(s)B_s{\rm d}s\;\;\;\operatorname P\text{-almost surely}\tag 1$$ for all $t\ge 0$? I've tried to obtain $(1)$ by the Itō formula in the following way: Let $$g(t,x):=f(t)x\;\;\;\text{for }t\ge 0\text{ and }x\in\mathbb R\;.$$ Then, $g\in C^{1,\:2}(\mathbb R_{\ge 0}\times\mathbb R)$ with $$\frac{\partial g}{\partial t}(t,x)=f'(t)x\;\;\;\text{and}\;\;\;\frac{\partial g}{\partial x}(t,x)=f(t)\tag 2$$ for all $t\ge 0$ and $x\in\mathbb R$. Thus, $$f(t)B_t=\int_0^tf(s)\;{\rm d}B_s+\int_0^tf'(s)B_s\;{\rm d}s\;\;\;\operatorname P\text{-almost surely}\tag 3$$ for all $t\ge 0$. Since we would need $f(t)B_t=0$ (i.e. $f(t)=0$) to obtain $(1)$ and since this needs to be the case for any $t\ge 0$, I start to think that $(1)$ is wrong. However, since Did wrote that $(1)$ is trivial for $f\in C^2(\mathbb R_{\ge 0})$, I probably made a mistake.

$\endgroup$
  • $\begingroup$ In my opinion the answer is no because Brownian Motion is almost everywhere not differentiable. But I'm no expert on the subject. $\endgroup$ – JonesY Jul 24 '16 at 18:52
  • 3
    $\begingroup$ Is your beef with the fact that $f$ is only $C^1$ and not $C^2$? This should be explained in the question... $\endgroup$ – Did Jul 24 '16 at 18:55
  • $\begingroup$ Why don't you address my specific request? Note that when $f$ is $C^2$, the identity is direct. $\endgroup$ – Did Jul 25 '16 at 10:07
  • $\begingroup$ Yeah it is: Itô. (As always...) $\endgroup$ – Did Jul 25 '16 at 10:13
  • $\begingroup$ @Did I've updated the question and deleted my obsolete comments. $\endgroup$ – 0xbadf00d Jul 25 '16 at 13:23
1
$\begingroup$

I doubt that this identity holds true. Just consider $f:=1$, then the assertion reads $$B_t = 0,$$ which is obviously not correct.


As @Did pointed out in a comment, Itô's formula shows that the identity $$\int_0^t f(s) \, dB_s = f(t) B_t - \int_0^t f'(s) B_s \, ds$$ holds for $f \in C^1(\mathbb{R})$.

$\endgroup$
  • $\begingroup$ I don't understand your answer. Why do you think we need $f\in C^2$? $(3)$ from the question holds for any $f\in C^1$ by the Itō formula. $\endgroup$ – 0xbadf00d Jul 26 '16 at 9:57
  • $\begingroup$ @0xbadf00d Since we are turning around this point for some time now, here is a suggestion: state precisely and completely the version of Itô's formula that you know and use -- then we might be able to have productive exchanges. $\endgroup$ – Did Jul 26 '16 at 11:43
  • $\begingroup$ @Did You can find it here. $\endgroup$ – 0xbadf00d Jul 26 '16 at 12:12
  • $\begingroup$ @0xbadf00d So, $f$ $C^1$ here yields $g:(t,x)\to f(t)x$ of class $C^{1,2}$ in your book, right? No mystery remaining, it seems. $\endgroup$ – Did Jul 26 '16 at 12:44
  • 1
    $\begingroup$ @0xbadf00d As Did pointed out, there are several versions of Itô's formula... and if you assume $f \in C^2$, you are, somehow, on the safe side. Anyway, you are right, it works fine for $f \in C^1$. $\endgroup$ – saz Jul 26 '16 at 13:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.