Let $U$ be a separable $\mathbb R$-Hilbert space and $W$ be a $Q$-Wiener process on a complete and right-continuous filtered probability space. Let $H$ be a separable $\mathbb R$-Hilbert space and $X$ be a $H$-valuedcontinuous adapted process with $$X_t=X_0+\int_0^t\varphi_s\:{\rm d}s+\int_0^t\Phi_s\:{\rm d}W_s$$ for some $H$-valued processes $\varphi$ and $\Phi$ such that the integrals are well-defined. The Itō formula, as presented by Da Prato and Zabczyk in Theorem 4.32 of their book Stochastic Equations in Infinite Dimensions, yields $$f(t,X_t)-f(0,X_0)=\int_0^t\frac{\partial f}{\partial t}(s,X_s)\:{\rm d}s+\frac12\int_0^t{\rm D}_x^2f(s,x_s)\:{\rm d}[\![X]\!]_s+\int_0^t{\rm D}_xf(s,X_s)\:{\rm d}X_s$$ for all $t$ almost surely for suitable $f:[0,\infty)\times H\to\mathbb R$. Above $[\![X]\!]$ denotes the tensor-quadratic variation of $X$. In the second integral, the second Fréchet derivative is considered as taking values in $(H\:\hat\otimes_\pi\:H)'$.

If $H=\mathbb R$, we're able to apply the Itō formula once again to obtain that $$Z_t:={\rm D}_xf(t,X_t)=\frac{\partial f}{\partial x}(t,X_t)\;\;\;\text{for }t\ge0$$ is again a semimartingale with $$[Z,X]_t=\left[\frac{\partial ^2f}{\partial x^2}(\;\cdot\;,X)\cdot X,X\right]_t=\int_0^t\frac{\partial ^2f}{\partial x^2}(s,X_s)\:{\rm d}[X]_s$$ for all $t$, where $[\;\cdot\;,\;\cdot\;]$ denotes the scalar-quadratic covariation and $\frac{\partial ^2f}{\partial x^2}(\;\cdot\;,X)\cdot X$ denotes the Itō integral process.

I would like to obtain a similar result in the General case. However, there are two major problems:

  1. How do we need to define the stochastic integral $\int_0^t{\rm D}_x^2f(s,X_s)\:{\rm d}X_s$? (The problem being that the integration theory as presented by Da Prato and Zabczyk only deals with Hilbert space valued integrals, but here the integrand is takes values in $\mathfrak L(H,\mathfrak L(H,\mathbb R))$)
  2. How do we need to define the tensor-quadratic covariation $[\![X,Z]\!]$? (The problem being that $Z$ is not Hilbert space valued, but $\mathfrak L(H,\mathbb R)$-valued; I know that we're in the special Situation where $\mathfrak L(H,\mathbb R)\cong H$, but I would like to come up with a solution which works for general operator-valued processes)

Last remark: I know that there is a stochastic integration theorey for Banach space valued processes. However, since above we're only dealing with processes which take values in spaces of operators between Hilbert spaces, I hope that there is an easy generalization available which is based on the Hilbert space valued theory.


Your Answer

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

Browse other questions tagged or ask your own question.