3
$\begingroup$

I've been working on the following question:

Let the vector $(X,Y,Z)$ be a multivariate Gaussian random variables with mean vector and covariance matrix:

$\mu = \begin{pmatrix} 0 \\ 3 \\ -1 \end{pmatrix}$ and $\Sigma = \begin{pmatrix} 16 & 8 & 0 \\ 8 & 9 & -2 \\ 0 & -2 & 1 \end{pmatrix}$

Derive the distribution of $(X,Z)$ given $Y=0$

I know (from my lecture notes) that the conditional distribution of $X_1$ given $X_2=x_2$ is Gaussian with mean: $\mu_1 + \Sigma_{12}\Sigma_{22}^{-1}(x_2-\mu_2)$ and variance-covariance matrix: $\Sigma_{11}-\Sigma_{12}\Sigma_{22}^{-1}\Sigma_{21}$

So I rewrote $\Sigma$ as $\Sigma = \begin{pmatrix} 16 & 0 & 8 \\ 0 & 1 & -2 \\ 8 & -2 & 9 \end{pmatrix}$

And then I let $X_1=\begin{pmatrix} X \\ Z \end{pmatrix}$ and $Y_2=(Y)$. So $E(X_1)=\begin{pmatrix} 0 \\-1 \end{pmatrix}$ and $E(X_2)=(3)$

And $Cov(\begin{pmatrix} Y_1 \\ Y_2 \end{pmatrix}) = \begin{pmatrix} \Sigma_{11} & \Sigma_{12} \\ \Sigma_{21} & \Sigma_{22} \end{pmatrix}$ where $\Sigma_{11}=\begin{pmatrix} 16 & 0 \\ 0 & 1 \end{pmatrix}, \Sigma_{12}=\begin{pmatrix} 8 \\ -2 \end{pmatrix} = \Sigma_{21}^T, \Sigma_{22}=(9)$

Meaning that $E(Y_1 \mid Y_2 =0) = \begin{pmatrix} 0 \\ -1 \end{pmatrix} + \begin{pmatrix} 8 \\ -2 \end{pmatrix}(\frac{1}{9})(0-3) = \begin{pmatrix} \frac{-8}{3} \\ \frac{1}{3}\end{pmatrix}$

and $Cov(Y_1 \mid Y_2=0) = \begin{pmatrix} 16 & 0 \\ 0 & 1 \end{pmatrix} - \begin{pmatrix} 8 \\ -2 \end{pmatrix}(\frac{1}{9})\begin{pmatrix} 8 & -2 \end{pmatrix} = \begin{pmatrix} 16 & 0 \\ 0 & 1 \end{pmatrix} - \begin{pmatrix} \frac{64}{9} & -\frac{16}{9} \\ -\frac{16}{9} & \frac{4}{9} \end{pmatrix} = \frac{1}{9}\begin{pmatrix} 80 & 16 \\ 16 & 5 \end{pmatrix}$

Am I correct in this approach? Is there something I'm fundamentally missing here? The subsequent question is "What is the distribution of $(2X-Z,3Y+Z)$?" How might I approach this one?

Thanks for any help/suggestions/feedback.

$\endgroup$
  • 1
    $\begingroup$ You might want to add the self-study tag and read it's wiki. $\endgroup$ – StubbornAtom May 12 '18 at 8:09
1
$\begingroup$

You have the correct formulas, but I leave it to you to check whether you've applied them correctly.

As for the distribution of $(2X−Z,3Y+Z)$, viewed as a 2 element column vector. Consider$(X.Y,Z)$ as a 3 element column vector. You need to determine the matrix $A$ such that $A*(X,Y,Z) = (2X−Z,3Y+Z)$ . Hint: what dimensions must $A$ have to transform a 3 by 1 vector into a 2 by 1 vector? Then use the result $\text{Cov} (A*(X,Y,Z)) = A* \text{Cov}(X,Y,Z)*A^T$ combined with the trivial calculation of the mean, and your knowledge of the type of distribution which a linear transformation of a Multivariate Gaussian has.

$\endgroup$
  • $\begingroup$ Thank you. Would you know any resource I can use to check if the first part is correct? $\endgroup$ – Ohmega May 12 '18 at 11:25
  • $\begingroup$ Your approach is correct, but you have at least 2 arithmetic errors in your calculations. These are not "deep" or conceptual errors, but careless errors. $\endgroup$ – Mark L. Stone May 12 '18 at 11:52

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.