When I retrace the implementation of lmer from lme4 I faced a question regarding cholesky decomposition used for solving penalized least squares.

Consider a Cholesky decomposition of a matrix M with block design

$ M = \left( \begin{matrix} A & B\\ C & D \end{matrix} \right) = RR'$

And the Cholesky desomposition described in the paper:

$\left( \begin{matrix} \Lambda_\theta^TZ^TWZ\Lambda_\theta + I & \Lambda_\theta^TZ^TWX\\X^TWZ\Lambda_\theta & X^TWX \end{matrix} \right) = \left( \begin{matrix} L_\theta & 0\\R^T_{ZX} & R_X^T \end{matrix} \right) \left( \begin{matrix} L^T_\theta & R_{ZX}\\0&R_X \end{matrix}\right)$

How can I extract $L_\theta$ from this decomposition?

Let's say

$\Lambda_\theta^TZ^TWZ\Lambda_\theta + I \in \mathbb{R}^{q\times q}$

$\Lambda_\theta^TZ^TWX \in \mathbb{R}^{q \times p}$

$X^TWX\in \mathbb{R}^{p \times p}$

I guess then follows that

$L_\theta \in \mathbb{R}^{q\times q}$ ,$R^T_{ZX} \in \mathbb{R}^{p \times q}$ and $R_X^T \in \mathbb{R}^{p \times p}$

My Idea is that $L_\theta$ must be the upper left block matrix of $R$, $R^T_{ZX}$ must be the lower block matrix of $R$ and so on. Am I right with my assumptions?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.