Re: Preconditioning and weighted least squares

From: spasmous (spasmous_at_yahoo.com)
Date: 03/16/05 

Date: 16 Mar 2005 11:33:51 -0800

Ray Koopman wrote:
>
> If the system is overdetermined then yes, you're doing weighted least
> squares, and the answer will depend on P. But if A is square and
> nonsingular then P should change nothing but roundoff error.

Thanks Ray, that now makes sense. So really that preconditioner should
come will a disclaimer or similar ;)

I've seen another popular type of preconditioner is to transform

A.x = b

into

A.P.x = b

This one looks statistically OK to me (ie. we are not weighting the
measurements in an arbitrary way).

Relevant Pages