2
$\begingroup$

Let

  • $(\Omega,\mathcal A,\operatorname P)$ be a probability space and $\lambda$ be the Lebesgue measure on $[0,\infty)$
  • $(H,\left\|\;\cdot\;\right\|)$ be a Banach space over the field $\mathbb F\in\left\{\mathbb R,\mathbb C\right\}$
  • $(X_t)_{t\ge 0}$ be a $H$-valued almost surely continuous stochastic process on $(\Omega,\mathcal A,\operatorname P)$

Let $\omega\in\Omega$ with $X(\omega)\in C^0([0,\infty);H)$. Then $$X_{[a,b]}(\omega)\text{ is compact in }(H,\left\|\;\cdot\;\right\|)\;\;\;\text{for all }0\le a\le b<\infty\;$$ Especially, $X(\omega)\in\mathcal L_{\text{loc}}^p(\lambda;H)$ and hence $$X(\omega)\varphi\in\mathcal L^p(\lambda;H)\;\;\;\text{for all }\varphi\in C_c^\infty([0,\infty);H)$$ for all $p\in [1,\infty)$. Thus, we can view $X(\omega)$ as being a distribution $$C_c^\infty([0,\infty);H)\to H\;,\;\;\;\varphi\mapsto\int X(\omega)\varphi\;{\rm d}\lambda\;.\tag 1$$

Is there anything I'm missing? Are some of my conclusions wrong? And what's the distributional derivative of $X$?

[I've read the Wikipedia article about the distributional derivative, but I don't know how we need to translate the definition to the setting described here].

EDIT: The question arose as I saw people talking about the distributional derivative of a (cylindrical) Brownian motion on a Hilbert space without giving a definition of what they mean. As @charlestoncrabb pointed out, my attempt in $(1)$ seems to be wrong, since the product $X(\omega)\varphi$ is not defined. Maybe we need to let $H$ be a Hilbert space with inner product $\langle\;\cdot\;,\;\cdot\;\rangle$ and replace $X(\omega)\varphi$ by $\langle X(\omega),\varphi\rangle$. By the same argumentation as before, the integral would exist. But honestly, I'm only guessing. Until now, I failed to generalize the usual notion to this scenario, but since people are using it, there must be some notion of distributional derivative of $X$.

$\endgroup$
1
$\begingroup$

The main issue I see here is that since $X,\varphi$ are $H-$valued, how do we define the product $X \varphi$? It seems you need additional structure to even get past this point (i.e., $H$ needs to be a Banach algebra). Assuming an algebra structure, how are you defining the space $\mathcal{L}^p(\lambda;H)$? In particular, what is the norm of this space? Finally, for a distributional derivative to make sense (by which I mean, align in some way with the usual usage), we need integration by parts to hold: i.e., "$Y$ is the distributional derivative of $X$ if": $$\int \nabla\varphi(\omega)X(\omega)d\lambda(\omega)=-\int\varphi(\omega)Y(\omega)d\lambda(\omega)$$ for all $\varphi\in C^\infty_c(\lambda;H)$ (noting here $\nabla$ refers to the Fréchet derivative). Again, for this to make sense, all of the prior issues need to make sense as well.

$\endgroup$
  • $\begingroup$ The question arose as I saw people talking about the distributional derivative of a (cylindrical) Brownian motion on a Hilbert space without giving a definition of what they mean. You're right, my attempt seems to be wrong. Maybe we need to let $H$ be a Hilbert space with inner product $\langle\;\cdot\;,\;\cdot\;\rangle$ and replace $X(\omega)\varphi$ by $\langle X(\omega),\varphi\rangle$. By the same argumentation as before, the integral would exist. $\endgroup$ – 0xbadf00d Jan 25 '16 at 13:01
  • $\begingroup$ But honestly, I'm only guessing. Until now, I failed to generalize the usual notion to this scenario, but since people are using it, there must be some notion of distributional derivative of $X$. $\endgroup$ – 0xbadf00d Jan 25 '16 at 13:01

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.