Reproducing results cross-platform

[Timo van Opstal]

I've downloaded precompiled binaries for Windows and Mac OS and run a simple geometry optimisation on a water molecule. However, the output differs significantlly, already the SCF iterations seem to behave better with the Windows binaries. I increased the verbose to inspect differences but can't find an apparent reason for the discrepancy between the two runs. Help is much appreciated!

CP2K version: 6.1, SVN revision 18464 

[Krack Matthias (PSI)]

Hi

 

Try again with a smaller EPS_DEFAULT value, e.g. 1.0E-12 instead of 1.0E-7, a larger CUTOFF value, at least 300 (Ry) instead of 200, and a larger REL_CUTOFF value, at least 40 instead of 30, to get numerically more meaningful results.

 

Matthias

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/748a2567-9315-4c6d-8b53-4cb2541c2614%40googlegroups.com.

[Timo van Opstal]

Thank you Matthias! The OSX version still misbehaves after these updates, I've also checked convergence in CUTOFF and REL_CUTOFF, (in the same vein as the tutorial) and it seems that my relative grid error is below 10^-4 in energy for those settings, so that should be OK I think?

On Sunday, September 29, 2019 at 3:49:50 PM UTC+2, Matthias Krack wrote:

Hi

 

Try again with a smaller EPS_DEFAULT value, e.g. 1.0E-12 instead of 1.0E-7, a larger CUTOFF value, at least 300 (Ry) instead of 200, and a larger REL_CUTOFF value, at least 40 instead of 30, to get numerically more meaningful results.

 

Matthias

 

Von: cp...@googlegroups.com <cp...@googlegroups.com> Im Auftrag von Timo van Opstal
Gesendet: Samstag, 28. September 2019 23:38
An: cp2k <cp...@googlegroups.com>
Betreff: [CP2K:12286] Reproducing results cross-platform

 

I've downloaded precompiled binaries for Windows and Mac OS and run a simple geometry optimisation on a water molecule. However, the output differs significantlly, already the SCF iterations seem to behave better with the Windows binaries. I increased the verbose to inspect differences but can't find an apparent reason for the discrepancy between the two runs. Help is much appreciated!

 

CP2K version: 6.1, SVN revision 18464 

--
You received this message because you are subscribed to the Google Groups "cp2k" group.

To unsubscribe from this group and stop receiving emails from it, send an email to cp...@googlegroups.com.

[Timo van Opstal]

This is still an open issue, but I'm working around it by avoiding the precompiled OS X binaries. Maybe there is an uninitialised variable somewhere.