Program gives floating point exception with Octave, not in MATLAB

[John W. Eaton]

On 16-Jul-1997, 0000-Admin(0000) <address@hidden> wrote:

| In trying to run the script included below on an DEC AlphaStation, we
| get the following error: 
| 
| Conjugate Gradient Method
| maximum number of iterations ---> [20,000]
| Error e ---> [1.e-01]
| dimension d ---> [100]
| number of patterns n ---> [10]
| 
| error: floating point exception -- trying to return to prompt

This is probably happening because you are generating Inf or NaN
somewhere in your script, and because gcc doesn't fully support IEEE
floating point arithmetic on the DEC Alpha.  The good news is that gcc
2.8.0 should support IEEE floating point arrithmetic on the DEC
Alpha.  The bad news is that no one knows when gcc 2.8.0 will be
released.

| The error occurs in different runs --- sometimes 6, sometimes 9. It
| doesn't happen when using MATLAB.

It probably occurs in different iterations because you are using data
generated by randn.

jwe