From help-octave-request  Thu May 27 14:24:23 1993
Subject: Re: sum(null matrix) becomes null matrix not zero 
From: John Eaton <jwe>
To: che6yoh at sun dot leeds dot ac dot uk
Cc: twang at meishan dot animal dot uiuc dot edu, help-octave
Date: Thu, 27 May 93 14:24:19 EDT

: > I think, octave uses eigenvectors to calculate expm, it could fail for 
: > some ill matrice.

Yes, Octave currently uses a fairly simple method based on eigenvalues
and eigenvectors to compute expm, logm, and sqrtm.  It's clearly not
sufficient, and needs to be improved.

: > However, MATLAB built-in expm function uses more general
: > and reliable algorithm.

Does anyone have a reference for the method used by Matlab?

I have a copy of Moler and Van Loan's paper `Nineteen dubious ways to
compute the exponential of a matrix', but it seems that there is some
kind of problem with every method, and that there is no clearly
superior way to do this computation.

Does anyone know of newer or better methods?

--
John W. Eaton      | The exam demonstrates a comminuted, slightly overlapping
jwe at che dot utexas dot edu|  | angulated fracture of the midfifth metatarsal.