I found it in Schilling, Partzsch's textbook "Brownian Motion":

only consider in $[0,T]$, they define the Dolean's measure $\mathbb P\times\mu$, and the corresponding $L^2$ norm on $L^2(\Omega\times [0,T],\mathcal F\times\mathcal B_T,\mathbb P\times \mu)$, where $\mu$ is the Lebesgue measure.

then define the elementary predictable set $\mathcal E_T$ and the corresponding stochastic integral.

then extend this definition to the closure of $\mathcal E_T$ under the $L^2$ norm above according to Ito isometry.

finally the author proved closure of $\mathcal E_T$ in $L^2(\Omega\times [0,T],\mathcal F\times\mathcal B_T,\mathbb P\times \mu)$ is $L^2(\Omega\times [0,T],\text{progressive sigma-algebra},\mathbb P\times \mu)$

my question is during the proof of the last statement, I don't find anything to do with the progressive property, the proof is valid if I replace it with "predictable sigma-algebra".

Do these two space coincide? I found that since

$\mathcal E_T\subset L^2(\Omega\times [0,T],\text{predictable},\mathbb P\times \mu)\subset L^2(\Omega\times [0,T],\text{progressive},\mathbb P\times \mu)$

take closure under $L^2$ norm, note that the second is a Hilbert space, we must have

$ L^2(\Omega\times [0,T],\text{predictable},\mathbb P\times \mu)=L^2(\Omega\times [0,T],\text{progressive},\mathbb P\times \mu)\tag{1}$

it seems strange, is there something wrong?

Does this also holds for any continuous square integrable martingale $M$?

GUESS: take

$A:=L^2(\Omega\times [0,T],\mathcal F\times\mathcal B_T,\mathbb P\times \mu_M)$

$B:= L^2(\Omega\times [0,T],\text{progressive},\mathbb P\times \mu_M)$

$C:=L^2(\Omega\times [0,T],\text{predictable},\mathbb P\times \mu_M)$

where $\mathbb P\times\mu_M$ is the corresponding Dolean's measure.

define $f_t=\langle M,M\rangle_t$

if $M$ is continuous square integrable martingale, and $f_t$ is absolutely continuous wrt the Lebesgue measure, then $A=B=C$ and the stochastic integral is defined for $C=B=A$ (motivated by the Shreve's textbook)

if $M$ is continuous square integrable martingale, then $C=B\subset A$ and the stochastic integral is defined for $C=B$

if $M$ is square integrable martingale, then $C\subset B\subset A$, and the stochastic integral is only defined for $C$

I found (1) i.e. (B=C) strange, because for any $f\in B$, (1) indicate that we can find $g\in C$ s.t. $$\mathbb E\left[\int_0^T|f_s-g_s|^2\,ds\right]=0$$

then $f=g$ a.s., that every progressive process is predictable? I think it is impossible.


For Brownian motion it is true that $B=C$. This means that if $f\in B$ there exists $g\in C$ such that $f(\omega,t)=g(\omega,t)$ for $\Bbb R\times\mu$-a.e. $(\omega,t)\in\Omega\times[0,T]$. This does not mean that $t\mapsto f(\cdot,t)$ is a predictable process. The matter is discussed in some detail in Chapter 3 of Introduction to Stochastic Integration by Chung and Williams.

  • $\begingroup$ thanks, the book you recommended is very useful! $\endgroup$ – Lookout May 9 '16 at 0:49

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.