3
$\begingroup$

Assume that $X_1$ and $X_2$ are correlated $N(0,1)$ variables. Now we can write \begin{align*} (X_1,X_2)^{T}=(\tilde{X_1},\gamma \tilde{X_1}+\sqrt{1-\gamma^2}\tilde{X_2})^{T} \end{align*}where $\tilde{X_1}$ and $\tilde{X_2}$ are independent $N(0,1)$ and $\gamma:=\operatorname{Cov}(X_1,X_2)$. I assume the same is possible for a longer vector. Does anybody know the general representation in terms of independent variables for vectors with $n$ components? This would be really helpful.

Thanks a lot.

$\endgroup$
3
$\begingroup$

Assume without loss of generality that the vector $X$ is centered and consider the nonnegative definite covariance matrix $C=E[XX^T]$. There exists a square matrix $L$ such that $LL^T=C$. Then, $X=LY$ where $Y$ is standard gaussian, that is, the entries of $Y$ are i.i.d. standard normal random variables.

Example: If $C=\begin{pmatrix}1 & \gamma \\ \gamma & 1\end{pmatrix}$, then a solution is $L=\begin{pmatrix}1 & 0 \\ \gamma & \delta\end{pmatrix}$ with $\delta=\sqrt{1-\gamma^2}$, which yields $$\begin{pmatrix}X_1 \\ X_2\end{pmatrix}=LY=\begin{pmatrix}Y_1 \\ \gamma Y_1+ \delta Y_2\end{pmatrix}.$$

See the WP page on the Cholesky decomposition and related algorithms.

$\endgroup$

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.