Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

I am interested to know if Ito integrals against Brownian motion can also be constructed via Skorohod representation. By this I mean the following: let $S_n$ be a simple random walk started at zero; for convenience assume the process linearly interpolates between times. Let $X^n$ be the process $$t \mapsto X^n_t = \frac{S_{nt}}{\sqrt{n}},$$ and of course the $X^n$ process converges weakly to Brownian motion (using the standard sup norm on $[0,1]$).

Now by Skorohod representation there exists a probabilty space supporting all the $X^n$ processes and a Brownian motion $B$ such that $X^n \to B$ almost surely. My question is if it is true that $$\int_0^1 f(t) dX^n_t \xrightarrow{a.s.} \int_0^1 f(t) dB_t$$ as $n \to \infty$, where $f$ is an $L^2$ function (deterministic even). The stochastic integral on the right is defined in the usual Ito type way as a limit in $L^2$ of the probability space.

It seems obvious that it is true for $f$ being an indicator function, and then by a density argument it goes through for general $f$. However I am a bit worried that I am missing something here and am looking for any insights. If it is true then that begs the question of whether or not it holds simultaneously for all $f$ at the same time, i.e. there is or is not an issue with null sets. Unfortunately null sets have a way of reducing me to a puddle of helplessness :)

ADDENDUM

Let me add a few more details to clarify a little and perhaps pique some interest. Skorohod representation tells us that there is a probability space $(\Omega, \mathcal{F}, P)$ and processes $X^n$ and $B$ defined on the space such that:

1) $t \mapsto X^n_t$ has the law of $t \mapsto S_{nt}/\sqrt{n}$,

2) $t \mapsto B_t$ is a Brownian motion,

3) the event $$E = \{ \omega \in \Omega : \sup_{0 \leq t \leq 1} |X_t^n(\omega) - B_t(\omega)| \to 0 \}$$ has $P(E) = 1$.

Note the first two statements are only about the marginal distribution of the processes, they say nothing about their joint distributions.

Now for any $f : [0,1] \to \mathbb{R}$ of the form $f(t) = \mathbf{1} \{ t \in I \}$, where $I$ is an interval of the form $[t_1, t_2], (t_1, t_2], [t_1, t_2), (t_1, t_2)$, and for any $\omega \in E$, one has $$ \int_0^1 f(t) dX_t^n (\omega) = X_{t_2}^n(\omega) - X_{t_1}^n(\omega) \to B_{t_2}(\omega) - B_{t_1}(\omega) = \int_0^1 f(t) dB_t(\omega) $$ as $n \to \infty$. Hence by linearity it follows that for all $f$ that are simple functions (finite linear combinations of indicator functions) and all $\omega \in E$ that $$ \int_0^1 f(t) dX_t^n(\omega) \to \int_0^1 f(t) dB_t(\omega). $$ The question is: to how large of a class of functions can the last equation be extended such that it holds for all $f$ in this class and all $\omega \in E$? It seems obvious that it should go beyond the class of simple functions, but I am not sure that it can go all the way up to $L^2([0,1])$.

share|improve this question
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Browse other questions tagged or ask your own question.