Re: Test fails - octave-2.1.20 on irix6.5

[Joao Cardoso]

"John W. Eaton" wrote:
> 
> On  8-Nov-1999, Joao Cardoso <address@hidden> wrote:
> 
> | "Billinghurst, David (RTD)" wrote:
> |
> | > Just built octave-2.1.20 on irix6.5 with both gcc-2.95.1 and egcs 19991102
> | > snapshot.
> | >
> | > I get a failure on test poly/residue-1.m as it misses the allowed 
> tolerances
> | >
> | > by an order of magnitude. Haven't done a full error analysis, but I 
> suspect
> | > the tolerances may be a little tight.
> | >
> | > The test case poly/residue-1.m is
> | >
> | > b = [1, 1, 1];
> | > a = [1, -5, 8, -4];
> | > [r, p, k, e] = residue (b, a);
> | > (abs (r - [-2; 7; 3]) < sqrt (eps)
> | >  && abs (p - [2; 2; 1]) < sqrt (eps)
> | >  && isempty (k)
> | >  && e == [1; 2; 1])
> | >
> | > For octave-2.1.20 this fails as
> | >
> | > r - [-2; 7; 3]  =
> | >    -1.0658e-14
> | >     1.7276e-07
> | >     1.1990e-14
> | >
> | > while sqrt (eps) =  1.4901e-08
> |
> | I also have this problem with SCO. However the same test in
> | octave-2.1.14 runs OK.
> 
> The roots function uses eig, and Lapack was updated to 3.0 for
> 2.1.20.  Could that be the cause of the problem?  It seems to work
> fine on my Linux system.
> 
> jwe

Yes, the problem is in eig().

In 2.1.29:

octave:15> A=[   5  -8   4;   1   0   0;   0   1   0]
A =
   
   5  -8   4
   1   0   0
   0   1   0
octave:16> r = eig (A)
r =
   
  2.00000
  2.00000
  1.00000
  
octave:17> r - [2 2 1]'
ans =

   -6.1055e-08
    6.1055e-08
   -4.4409e-16

while in 2.1.14

octave:47> r
r =

  2.00000 + 0.00000i
  2.00000 - 0.00000i
  1.00000 + 0.00000i
   
octave:48> p - [2 2 1]'
ans =

  -6.6613e-16 + 5.4657e-08i
  -6.6613e-16 - 5.4657e-08i
  -7.7716e-16 + 0.0000e+00i

Matlab on an alpha with OSF1:

bash-2.01$ matlab

                              < M A T L A B >
                  Copyright 1984-1999 The MathWorks, Inc.
                        Version 5.3.1.29215a (R11.1)
                                Oct  6 1999

 
  To get started, type one of these: helpwin, helpdesk, or demo.
  For product information, type tour or visit www.mathworks.com.
 
>> A=[   5  -8   4;   1   0   0;   0   1   0]
>> r = eig (A)

r =

   2.0000 + 0.0000i
   2.0000 - 0.0000i
   1.0000          

>> r - [2 2 1]'

ans =

   1.0e-07 *

   0.0000 + 0.2300i
   0.0000 - 0.2300i
   0.0000          


And now my knowledge has finish... can this help?


Joao

-- 
Joao Cardoso, address@hidden

"Science is like sex: sometimes something useful comes out, but that is
not the reason we are doing it" -- Richard Feynman