S.O.S. Mathematics CyberBoard

Hello,

I am trying to invert a large symmetrical covariance matrix (91, 91), and having a bit of a nightmare with it! I have tried using singular value decomposition, and whilst the product of A*A^(1) is approximately a unit matrix, the diagonal elements of the inverted matrix are a mixture of positive and negative values, whilst the diagonal elements of the original matrix are all positive, as such when I try to take the square root of the inverted matrix all hell breaks loose! I think that it may have something to do with the fact that the condition number is quite large. However I have some matrices that do not have a particularly large condition number (i.e. about 1000), and I am still running into problems. I am trying to do this in IDL by the way, and would appreciate any help what-so-ever.
Thanks,
Sam

Welcome to the CyberBoard, sam.illingworth! Please post your question only once. Thanks!

_________________
The greater danger for most of us lies not in setting our aim too high and falling short; but in setting our aim too low, and achieving our mark. - Michelangelo Buonarroti

I'm not really sure what your question is. However you cannot take the sqrt of a negative number. I would think this is why you run in to trouble

_________________
Rud Faden