SCM

Forum: Bug found and squished in octave-original code

Posted by: Martin Maechler
Date: 2008-02-29 23:44
Summary:Bug found and squished in octave-original code
Project:expm - Matrix exponential

Thanks to the R version of the "s.P.s" (scaling + Pade + squaring) algorithm,
from reading Ward (1977) and understanding his two extra pre-conditioning steps,
then re-doing them in R {via expm(., method="R_Ward77")}, in testing making heavy use of the R interface to LAPACK's DGEBAL,
I've finally been able to replace the buggy back-permutation by correct code, in a 2nd step doing the row- and column- back-permutation via BLAS 'DSWAP()'.

Now the plan is to make more extensive accuracy and speed comparisons.
Additionally I also want to implement a slightly different way of step 2, using DGEBAL("B", ..) instead of first "P" and then "S"

Latest News

 

Monitor Forum | Start New Thread Start New Thread
Thanks to:
Vienna University of Economics and Business University of Wisconsin - Madison Powered By FusionForge