[Octave-bug-tracker] [bug #37336] Incorrect result of rcond with Netlib LAPACK

[Sébastien Villemot]

URL:
  <http://savannah.gnu.org/bugs/?37336>

                 Summary: Incorrect result of rcond with Netlib LAPACK
                 Project: GNU Octave
            Submitted by: svillemot
            Submitted on: mer. 12 sept. 2012 16:19:10 GMT
                Category: Libraries
                Severity: 3 - Normal
                Priority: 5 - Normal
              Item Group: Incorrect Result
                  Status: None
             Assigned to: None
         Originator Name: 
        Originator Email: 
             Open/Closed: Open
         Discussion Lock: Any
                 Release: 3.6.2
        Operating System: GNU/Linux

    _______________________________________________________

Details:

I attach a MAT-file containing a 2x2 matrix (Finf), whose rcond is different
depending on the flavor of LAPACK that I use.

With Netlib LAPACK:

 octave:2> rcond(Finf)
 ans =  0.40463

With Atlas' LAPACK:

 octave:2> rcond(Finf)
 ans = 0

MATLAB gives 5.5511e-17.

Since the matrix is quasi-singular, the right answer is (almost) zero. A value
of 0.4 is plain wrong.

I cannot determine if this is a bug in Netlib LAPACK or Octave. I wrote a C
program reading the matrix with libmatio then calling dgetrf+dgecon and I get
the right result with both versions of LAPACK, so this points towards a bug in
Octave, but I am still unsure.




    _______________________________________________________

File Attachments:


-------------------------------------------------------
Date: mer. 12 sept. 2012 16:19:10 GMT  Name: burp.mat  Size: 216 o   By:
svillemot

<http://savannah.gnu.org/bugs/download.php?file_id=26542>

    _______________________________________________________

Reply to this item at:

  <http://savannah.gnu.org/bugs/?37336>

_______________________________________________
  Message posté via/par Savannah
  http://savannah.gnu.org/