# Covariance of Wiener Process


(...) and $W$ is a Wiener process with the $W_t$ jointly Gaussian, $W_0=0$, conditional means $\mathrm{E}(W_t\mid W_s)=W_s$ for $s\leq t$, which implies $\Cov(W_s,W_t)=\abs{t-s}$.

However, I managed to obtain the following: \begin{align} \Cov(W_s,W_t)&=\Cov(W_s-W_0,W_t-W_s+W_s-W_0)\\ &=\Cov(W_s-W_0,W_s-W_0)+\Cov(W_s-W_0,W_t-W_s)\\ &=s+0 \end{align}

Am I doing something wrong? What is it? How can I get the quoted result?

Thanks for helping! :D

• Which text is this? Are you sure $W_t$ in this snippet refers to the Wiener process? The quoted covariance is not correct (consider what happens when $t=s$). – Juho Kokkala Aug 19 '16 at 6:56
• Hey @JuhoKokkala I updated the question with the remaining part of the text (it comes from a lecture pdf). I think it is clear now that it refers to the Wiener process. What do you think? – Guilherme Salomé Aug 19 '16 at 15:44

There is an error in the quoted text. This becomes obvious when one considers $t=s$: $$\mathrm{Var}(W_t) = \mathrm{Cov}(W_t,W_t) "=" |t-t| = 0,$$ but the Wiener process does not have zero variance. Any other process cannot have that covariance function, either, since if the variances of $W_t$ and $W_s$ are zero, then their covariance must be zero, too.

The calculation in the question is correct if $s\leq t$ (as was assumed).

I'm trying to discern how the book made this error. In general $\Cov(W_s,W_t)=\min(s,t)$ for arbitrary $s,t\geq 0$, and if you recall the nifty formula:

$$\min(s,t)=\frac{s+t-|s-t|}{2},$$

then

$$\Cov(W_s,W_t)=\frac{s+t-|s-t|}{2}.$$

• Thanks for pointing that out. The original pdf seems to be missing a big chunk of the equation in this case. – Guilherme Salomé Aug 19 '16 at 18:56