energy rose up in a re-do calculation

Message

murphy · #1 Post by **murphy** » Tue Sep 24, 2013 11:41 am

Dear VASP users,

I am doing a calculation of CdSe cell containing ~100 atoms. After a ISIF=3 optimization, I re-did the calculation after copying CONTCAR to POSCAR, because I wanted to check if it was really converged (for this calculation, I set NPAR=4, and one more digit for EDIFF and EDIFFG). And even I did not keep the CHGCAR and WAVECAR, the total energy was always higher than previously converged value for ~0.2 eV.

Please explain for me why this happens. Thank you in advance!

Best,
Murphy

alex · #2 Post by **alex** » Tue Sep 24, 2013 1:51 pm

The trap you took is named Pulay stress.

Cheers,

alex

murphy · #3 Post by **murphy** » Thu Sep 26, 2013 11:17 am

Hi Alex,

You mean Pullay stress? Could you explain more?
Thanks!

Best,
Murphy

alicenancy · #4 Post by **alicenancy** » Wed Oct 23, 2013 7:40 am

Now, i was doing some test calculations to test the performance of the machine and I found out that the same job running in Virtual Node mode (using all 4 cores of the processors) takes 4 times wall clock time than the same job running in the Symmetrical Multi-Processing mode (only one task per node). Is it something expected? Is it possible to improve the performace of the VN mode somehow (using different libraries..)? Does anyone has experience in setting the NPAR parameter for those kind of machines?

murphy · #5 Post by **murphy** » Wed Nov 06, 2013 5:02 pm

Eventually the problem was solved by setting EDIFF=10^-5, and redo ISIF=3 using the CONTCAR of the last ISIF=2 relaxation. This problem should be due to the fact that the ISIF=3 relaxation changes the shape of the supercell, hence the number of wavefunctions in x, y, and z directions in the reciprocal space. This will result in energy jump. Therefore I can not run a ISIF=3 relaxation using a CONTCAR obtained from another ISIF=3 calculation.