[Ipopt] Using ipopt with PARDISO

[Andreas Waechter]

Hi,

>> Anyway, for now I just wanted to compare ipopt with MA27 vs. ipopt
>> with PARDISO in terms of robustness. I suppose the optimized BLAS/LAPACK
>> don't change the results, just the CPU time, am I right?
>
> The BLAS shouldn't make a difference. But at least for gcc 4.3 on
> Linux 64bit

Different implementations and compilations of BLAS or LAPACK (or any 
numeric code) can lead to having floating point operations done in a 
different order (particularly when we talk about blocking in BLAS).  As a 
consequence, the results can be slightly different in the very last 
digits.  Unfortunately, for large and numerically unstable problems, this 
can lead to different behavior of Ipopt.

Regards,

Andreas