Comparing serial to parallel version

Message

mtvand2 · #1 Post by **mtvand2** » Tue Apr 28, 2009 10:48 pm

I am interested in learning how I might confirm that my parallel version of VASP is working properly. Specifically, I'm concerned because I get a different final energy when I run the same input on a different number of processors. See the following:

NProc F E0 d E
1
-.74924471E+03 -.74855157E+03 -.165608E+01
2
-.74924477E+03 -.74855118E+03 -.165607E+01
4
-.74924458E+03 -.74855221E+03 -.165622E+01
16
-.74924626E+03 -.74855516E+03 -.166517E+01
32
-.74924372E+03 -.74855286E+03 -.164182E+01

The more processors, the further from the single-processor energy. There must be a standard input that I can use to test my installation (to compare my output to a standard output). I have searched the VASP forum, but have been unable to find this information if it is out there.

alex · #2 Post by **alex** » Thu Apr 30, 2009 9:05 am

Hi,

I would guess you have an convergency issue there. Set EDIFF to a rather small value (e.g. 1.e-8 or so) and try again. The differencies should not exceed micro-eV in the total energy.

One other attempt might be to compare the very first line of numbers in OSZICAR. If they are the same and not the converged energy, where numerical noise might already had impact on, you should be on the safe side.

Just in case: Did you compare single point energies?

Cheers

Alex