EDIT: Showing this using Ito's lemma is easy, that's NOT what I want to do. I also realised that $2\mathbb{E}[S\xi]\neq 2\xi\mathbb{E}[S]$ since $\xi$ is also a random variable. Nontheless, if this is the case I have no idea how to calculate the expectation of $S\xi$ anyway.

Given a Brownian motion $W(t)$ I want to show that

\begin{align} \lim_{n\rightarrow\infty}\mathbb{E}\left[\left|\sum_{j=0}^{n-1}\frac{jT}{n}\left(W\left(\frac{(j+1)T}{n}\right)-W\left(\frac{jT}{n}\right)\right) - TW(T)+\int\limits_0^TW(t)dt\right|^2\right]=0 \tag1. \end{align}

For simplicity we denote the sum by $S$ and set $\xi=-TW(t)+\int_{0}^TW(t) \ dt$, now we have that \begin{align} \mathbb{E}[(S+\xi)^2]=\mathbb{E}[S^2]+2\xi\mathbb{E}[S]+\xi^2. \end{align}

However I have trouble calculating $\mathbb{E}[S^2]$. I know that $\mathbb{E}[S]=0$ since

\begin{equation} \mathbb{E}[S]=\sum_{j=0}^{n-1}\frac{jT}{n}\mathbb{E}[W_{j+1}-W_j]=0 \end{equation} since the increments are $\sim\mathcal{N}(0,T/n)$ so the sum above is just a sum of zeroes. So I'm kind of left to show that

\begin{equation} \lim_{n\rightarrow\infty}\mathbb{E}[S^2]=-\xi^2. \end{equation}

I tried the following: For Brownian motion I know that $\mathbb{E}[S^2]=\text{Var}[S]$ so

\begin{align} \mathbb{E}[S^2]&=\sum_{j=0}^{n-1}\frac{j^2T^2}{n^2}\text{Var}[W_{j+1}-W_j]=\sum_{j=0}^{n-1}\frac{j^2T^2}{n^2}\frac{T}{n}=\frac{T^3}{n^3}\sum_{j=0}^{n-1}j^2\\&={\frac {{T}^{3} \left( 2\,{n}^{2}-3\,n+1 \right) }{6{n}^{2}}} \end{align}

which gives $T^3/3$ when $n\rightarrow\infty$. But this is not the result I want.

So you might ask how did I end up in (1)? Well, I wanted to show, using the definition of the Ito integral that

$$\int_0^TW(t)dt+\int_0^TtdW(t) = TW(T).$$

I rewrote it as

$$\int_0^TtdW(t) = TW(T) - \int_0^TW(t)dt$$

and used the definition: If there exists a stochastic process $I(T)$ such that $||I_n(T)-I(T)||_2=\lim\limits_{n\rightarrow\infty}\mathbb{E}[|I_n(T)-I(T)|^2]=0$ then $I(T)$ is an Ito integral. Plugging in $I_n(T)$ and $I(T)$ I arrived to (1).

  • 2
    $\begingroup$ What you can do is to discretize $TW(T)$ and $\int_0^TW(t)dt$ using the same partition, and then combine all them together to reach the term $-\frac{T}{n}W(T)$. $\endgroup$
    – Gordon
    Feb 1, 2021 at 22:05
  • $\begingroup$ Thanks a lot, I will try this tomorrow! $\endgroup$
    – Parseval
    Feb 1, 2021 at 22:59
  • $\begingroup$ @Gordon Is it correct partitioning the integral as $$\int_0^{T}W(t)dt\approx\sum_{0}^{n-1}\frac{jT}{n}W\left(\frac{jT}{n}\right) \ ?$$ For $TW(T)$, does the partition simply become the same as above? $\endgroup$
    – Parseval
    Feb 2, 2021 at 8:47
  • $\begingroup$ If I only partition $\int_0^TW(t)dt$ I obtain 0 in the expectation. I leave $-TW(T)$ as is then I get the combined sum in the argument of the expectation to be only $TW(T)$. Can this be correct? $\endgroup$
    – Parseval
    Feb 2, 2021 at 9:27
  • $\begingroup$ $W(T)=\sum_{j=0}^{n-1} \big(W(\frac{(j+1)T}{n})-W(\frac{jT}{n})\big)$ and $\int_0^TW(t)dt\approx \sum_{j=0}^{n-1}\frac{T}{n}W(\frac{jT}{n})$. $\endgroup$
    – Gordon
    Feb 2, 2021 at 12:36

1 Answer 1


Note that \begin{align*} \int_0^T W(t)dt \approx \sum_{j=0}^{n-1}\frac{T}{n}W\Big(\frac{jT}{n}\Big). \end{align*} Then, \begin{align*} &\ \sum_{j=0}^{n-1}\frac{jT}{n}\bigg(W\Big(\frac{(j+1)T}{n}\Big)-W\Big(\frac{jT}{n}\Big)\bigg) - TW(T) + \int_0^T W(t)dt\\ \approx &\ \sum_{j=0}^{n-1}\frac{jT}{n}\bigg(W\Big(\frac{(j+1)T}{n}\Big)-W\Big(\frac{jT}{n}\Big)\bigg) - TW(T) + \sum_{j=0}^{n-1}\frac{T}{n}W\Big(\frac{jT}{n}\Big)\\ =&\ \sum_{j=0}^{n-1}\frac{jT}{n}W\Big(\frac{(j+1)T}{n}\Big) - \sum_{j=0}^{n-1}\frac{(j-1)T}{n}W\Big(\frac{jT}{n}\Big) - TW(T)\\ =&\ \sum_{j=0}^{n-1}\frac{jT}{n}W\Big(\frac{(j+1)T}{n}\Big) - \sum_{j=-1}^{n-2}\frac{jT}{n}W\Big(\frac{(j+1)T}{n}\Big) - TW(T)\\ =&\ \frac{(n-1)T}{n}W(T) - TW(T)\\ =&\ -\frac{T}{n} W(T). \end{align*}

  • $\begingroup$ Ah, alright, so you did not actually discretize $TW(T)$ but only approximated the integral with the Riemann sum. But how did you combine the last two sums, when they differ in indexes? I assume that what you've done is this: $$\frac{T}{n}\left(\sum_{j=0}^{n-1}W\left(\frac{(j+1)T}{n}\right)-\sum_{j=-1}^{n-2}W\left(\frac{jT}{n}\right)\right)=\frac{T}{n}\sum_{j=0}^{n-1}\left(W\left(\frac{(j+1)T}{n}\right)-W\left(\frac{jT}{n}\right)\right)=\frac{T}{n}W(T)$$ $\endgroup$
    – Parseval
    Feb 2, 2021 at 17:07
  • $\begingroup$ I missed the $j$ factor above that multiplies the $W'$s. What happens to them? Things are clear till the 4th line, I don't see how you go from 4th to 5th. $\endgroup$
    – Parseval
    Feb 2, 2021 at 17:22
  • $\begingroup$ I assume he did the following: Remove the last term on the left hand sum, the first term on the right hand sum. Then you get two sums from $j =0$ to $n-2$. Combine those sums, and then add back in the terms you popped off separately. $\endgroup$ Feb 2, 2021 at 17:34
  • 1
    $\begingroup$ @rubikscube09- what do you mean by "remove the last term"? I can't just remove it because it will change the value of the expression. EDIT: Ok I think I understand what you mean now. $\endgroup$
    – Parseval
    Feb 2, 2021 at 17:42

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.