2
$\begingroup$

I have a quadratic form of the form $Q(\Sigma; x, \mu) = (x-\mu)'\Sigma^{-1}(x-\mu)$ where $\Sigma$ is a positive-definite non-singular matrix with (modified) Cholesky decomposition $\Sigma = LDL'$ with $L$ being a lower-triangular matrix (with diagonals all unity).

I want to find the derivative of the above quadratic form with respect to the entries in $L$. That is, I want to find

$$\frac\partial{\partial L} Q(\Sigma;x,\mu).$$

Is this an established result? are there references I could look at?

Many thanks!

$\endgroup$
5
  • $\begingroup$ you might find the answer here math.uwaterloo.ca/~hwolkowi/matrixcookbook.pdf $\endgroup$ Jul 27, 2015 at 15:21
  • $\begingroup$ I have looked at this book. The derivatives are only wrt unconstrained matrices. Do the constraints (of lower triangularity of the matrix) not matter? $\endgroup$ Jul 27, 2015 at 15:41
  • $\begingroup$ I guess I am a bit confused also -- the vech operator (defined for the lower-triangle) should apply. But that is typically defined for symmetric matrices. Does this matter that this is not a symmetric matrix but we are only interested in one triangle (and the other triangle is a constant zero). $\endgroup$ Jul 27, 2015 at 15:53
  • $\begingroup$ I would suggest you should try out with some 2X2 matrices first and from there prove the formula --that how we learn maths :) $\endgroup$ Jul 27, 2015 at 16:15
  • $\begingroup$ So, I guess try and reinvent the wheel is what you would suggest as the best option. I was planning to do that, but i wanted to know if there were standard results already available. $\endgroup$ Jul 27, 2015 at 16:46

0

You must log in to answer this question.

Browse other questions tagged .