Suppose that $X=(X_t)_{t \in [0,1]}$ is a continuous Gaussian process, for which $\mathbb{E}(X_t) = 0$ for all $ t \in [0,1]$ and $Cov(X_s,X_t) = s(1-t)$ for all $0 \leq s \leq t \leq 1 $. Let $Y \sim N(0,1)$ be a random variable independent of $X$. Define the stochastic process $W = (W_t)_{t \in [0,1]}$ by setting $W_t = X_t + tY$.

Show that the conditional distribution of the process $W$ given $\{W_1 =y\}$ is Gaussian, and calculate their mean and covariance functions. As a hint it is given how to make sense of conditioning on a set of zero probability:

The event $\{W_1 =y \}$ has zero probability, but it is natural to define conditioning on this event by the following limiting procedure. For any $\varepsilon >0$, the event $\{|W_1 - y| < \varepsilon\}$ has positive probability, so we can consider the conditional distribution of the process $W$ given the event $\{|W_1 - y| < \varepsilon\}$, and then take the (weak) limit of this conditional distribution as $\varepsilon \downarrow 0$.

I have already shown that W is a Gaussian process. Now my idea was to calculate $\frac{\mathbb{P}(\{W \in B\} \cap \{|W_1 - y| < \varepsilon\})}{\mathbb{P} \{|W_1 - y| < \varepsilon\}}$, but I do not seem to come to the right conclusion. Is this even the right approach? Further I do not see where I have to use weak convergence here. Thank you very much in advance for any help.


You can finesse this by showing that the distribution of $(X_t+ty)_{t\in[0,1]}$ as $y$ varies over $\Bbb R$, (call it $Q_y$) is a regular conditional distribution of $W$ given $W_1$. Namely, $y\mapsto Q_y$ is suitably measurable and $$ P[W\in B, W_1\in C]=\int_{C}Q_y(B)\varphi(y)\,dy,\qquad(*) $$ where $\varphi$ is the standard normal density. As $X_t+ty$ is clearly Gaussian, this will meet your needs. To see (*), just compute: Becaue $Y$ is independent of $X$, $$ \eqalign{ P[W\in B, W_1\in C] &=P[(X_t+tY)_{t\in[0,1]}\in B, Y\in C]\cr &=\int_C P[(X_t+ty)_{t\in[0,1]}\in B]\cdot P[Y\in dy]\cr } $$

  • $\begingroup$ Thank you very much for your answer. There are a few things that are not clear to me. If I do it as you said, I would get somthing like $\frac{\int_{-\varepsilon+y \leq Y \leq \varepsilon +y} Q_y(B) \varphi(y) \ dy}{\mathbb{P}[\{-\varepsilon + y \leq Y \leq \varepsilon y\}]}$. Wouldn't this blow up as $\varepsilon \downarrow 0$. Further I still do not understand where weak convergence is used here. $\endgroup$ – Rupert R Mar 24 at 8:09

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.