4
$\begingroup$

Is the following inequality true? I can't find a counterexample so I start to believe it is true but I do not manage to prove it :) Any ideas?

Let $f$ be a compactly supported bounded twice continuously differentiable function with bounded derivatives, then $$\int_0^t \int_{\mathbb{R}} f''(z) \frac{1}{\sqrt{2\pi s}} e^{- \frac{z^2}{2s}}dz ds \leq C [f]_{\alpha}\leq C [f]_{\alpha} + C\|f\|_{\infty} = C \|f\|_{\alpha}$$ for some finite constant $C>0$ where $$[f]_\alpha := \sup_{x\neq y} \frac{|f(x)-f(y)|}{|x-y|^{\alpha}}$$and here $$\|f\|_{\alpha} := [f]_{\alpha} + \|f\|_{\infty}$$ is the Hölder norm.

Observation: one can upperbound the integral by $C\|f'\|_{\infty}$ by a simple use of integration by parts. Using integration by parts twice is too much, the time integral explodes, so the question remains if there is a middle step in order to obtain the Hölder norm in the right hand side.

Thanks!

$\endgroup$
  • 2
    $\begingroup$ I am confused, do you mean $f''(z)$? $\endgroup$ – Lost1 Oct 23 '15 at 20:23
  • $\begingroup$ Yes sorry. Changed. $\endgroup$ – Martingalo Oct 23 '15 at 20:25
  • $\begingroup$ And what should the $\int_0^t$ at the start really be? $\endgroup$ – David C. Ullrich Oct 23 '15 at 20:38
5
$\begingroup$

Let $W$ be the standard Wiener process. Then, by the Ito formula $$ E[f(W_t) -f(0)] = E\left[\int_0^t f'(W_s)dW_s\right] + \frac12 E\left[\int_0^t f''(W_s)ds \right]\\ = \frac12\int_0^t \int_{\mathbb{R}} f''(z) \frac{1}{\sqrt{2\pi s}} e^{- \frac{z^2}{2s}}dz\, ds, $$ whence $$ \left|\int_0^t \int_{\mathbb{R}} f''(z) \frac{1}{\sqrt{2\pi s}} e^{- \frac{z^2}{2s}}dz\, ds\right| \le 2 E\big[|f(W_t)-f(0)|\big]\le 4||f||_\infty. $$

So even the supremum norm alone is enough to bound the integral.

$\endgroup$
  • $\begingroup$ This is what I was in the process of typing out, except I think the top line should be $F(W_t)-F(0)$? $\endgroup$ – Lost1 Oct 23 '15 at 20:41
  • $\begingroup$ @Lost1, sorry, it should be $f(W_t) - f(0)$ (it is already). $\endgroup$ – zhoraster Oct 23 '15 at 20:42
  • $\begingroup$ Oh... Does this actually imply that one can upperbound $\int_0^t \int_{\mathbb{R}} f^{(n)}(z) \frac{1}{\sqrt{2\pi s}}\exp (-z^2/(2s) ) dzds$ for any $n$-th derivative using the argument above iteratively? $\endgroup$ – Martingalo Oct 23 '15 at 20:53
  • $\begingroup$ @Martingalo, no. $\endgroup$ – zhoraster Oct 23 '15 at 20:53
  • $\begingroup$ @Martingalo, but partially, yes. You can integrate $E[f^{(n-2)}(W_t)] = \int_{\mathbb{R}} f^{(n-2)}(\sqrt{t}x) \varphi(x) dx$ by parts many times. You will end up with an estimate like $|f^{(n-2)}(0)| + C(t) ||f||_\infty$. Unfortunately, $C(t)$ explodes as $t\to 0+$. $\endgroup$ – zhoraster Oct 23 '15 at 20:59

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.