Re: pseudoinverse

"Michael A. Koerber" <[email protected]> Fri, 22 Nov 2002 08:32:04 -0500
Newsgroups gmane.lisp.matlisp.user
Message-ID <[email protected]>
Tunc,

> When M<N, should the matrix have rank=M or does it
> not matter?

1.  Good question.  The answer is, PINV as written will fail or
    produce a numerically unstable result due to taking the reciprocal
    of a very small singular value.

2.  The solution is straight forward, let all singular values < MACH-TOL
    equal zero.  ==> rank(A) = r <= min(M,N).  And use just those 
    r elements to form PINV.

3.  I have an all day meeting today, but will make these changes in the
    next couple of days.  ....BTW what is the name of the "*MACH" in 
    the matlisp core....

more later


mike